AI transformationof your businessAI transformation of your business

Overview

Iskra is an artificial intelligence platform for personal and team work. It brings together an AI chat, an agent mode for complex multi-step tasks, work with documents and data, Memory, skills, and projects for collaboration.

Overview

This section helps you sign in, create an account, and complete the initial setup of the Iskra platform.

Signing in

To sign in, open the Iskra link provided by your administrator. On the sign-in screen, you need to enter:

Login or email — the email address you used when signing up. The field accepts either an email address or a nickname. To sign in with a nickname, use Latin letters, digits, hyphens, or underscores. Enter your email address in full, exactly as you did when signing up. If the field contains characters from another alphabet, they are kept, and a hint appears below the field explaining how the rules for nicknames and email addresses differ.

Password — your account password.

Additional options on the sign-in screen:

Remember me — this checkbox is selected by default: you stay signed in for up to 30 days, and the period is extended as you use Iskra. After a break of two to three weeks, you will not need to enter your password again. Clear the checkbox on someone else’s computer: your sign-in will then be short-lived, 24 hours at most, and closing the browser may also end it.

Forgot password? — a link for regaining access (see “Password recovery” below for details).

Language switch. On the sign-in, sign-up, and access recovery screens, you can choose Russian, English, French, or Simplified Chinese. Your choice is saved in the current browser and applied to the interface. Messages, documents, and other user content are not translated automatically. After you sign in, the interface language and the response language are configured separately under Settings → App.

After you sign in successfully, you land on the main screen with the chat. If your email address has not been verified yet, the system redirects you to the verification screen.

Main screen

After you sign in, Iskra opens in its unified workspace interface:

Left panel — the main navigation: new chat, projects, Memory, skills, schedule, chat search, archive, and profile settings. If apps are enabled on your platform, an “Apps” section also appears here for viewing and managing them (see the “Apps” chapter).

Central work area — the current section. Chat, Memory, projects, schedule, and skills all use the same working width, so content does not “jump” in size when you switch between sections.

Top buttons — quick actions for the current screen. The chat title and the “⋯” (three dots) button at the top left open the chat actions menu: chat statistics, sharing, and downloading the chat archive (it contains the entire chat, including the response timelines and files), and below the divider — “Star”, “Rename conversation”, “Move to organization”, and “Move to archive” (the same items as in the menu of a chat row in the left panel, except that there the groups are in reverse order: “Star” and the chat actions on top, and statistics, “Share”, and archiving below the divider). On the right are the profile/organization switcher (an avatar, a name, and a chevron; clicking it opens the list of profiles) and notifications; if you do not have a paid subscription, a green “More features” badge to the left of the switcher leads to the pricing page. The header no longer has a limit indicator — monthly limit usage is shown in the profile menu at the bottom of the sidebar, in the “Usage” item (the percentage used is shown on the right, as on the statistics page; profiles without a monthly limit do not have this item); clicking it opens the settings at the “Statistics” section. The profile switcher is also available in the profile menu at the bottom of the sidebar: click your name, then the first item, “Profile”. Further down this menu are “User”, “Usage”, “Settings”, and “Help” (which contains the help, keyboard shortcuts, “Feedback”, and documents), and below the divider — “Sign out”. On mobile, the switcher (a circle and the workspace name, shortened if it is long) is on the right of the header, before the notifications and the “⋮” (three dots) button, which opens a bottom sheet with the chat title, its actions, and the “Archive” item. On mobile, you create a new chat from the side menu.

Right panel — in a chat, this is the “request panel”: it shows what the next request will be handled with (connected Memory folders and chat files), and its bottom row shows how full the context is, buttons for conversation skills and pinned notes, and the response parameters. On a project page, the right side shows the project materials. You can collapse the panel if you need more room for the central area.

Compact icon buttons in the chat, navigation, chat files, schedule, and projects have text tooltips. Hover over a button or move keyboard focus to it to see the name of the action.

The background and spacing of the central area match the side panels in both the light and dark themes. You can switch the theme in the app settings.

Signing up

If your administrator has enabled self-registration, the sign-in screen has a “Create account” link. Click it to go to the sign-up form.

The sign-up form contains the following fields:

First name (required) — your real first name.

Last name (optional) — your last name.

Nickname (required) — a unique username. The system automatically generates a nickname from your first and last name (by transliteration). You can edit it manually. Allowed characters: Latin letters, digits, hyphens, and underscores. Length: 3 to 32 characters.

Email (required) — the email address you will use to sign in and to recover your password.

Password (required) — choose a strong password. A password strength indicator below the field visually shows how secure the chosen password is.

Confirm password (required) — enter the password again. If the passwords do not match, you will see a warning.

After filling in all the fields, click “Create account”. If sign-up succeeds, you will be redirected to the email verification screen.

Email verification

After signing up (or when you try to sign in with an unverified email address), you land on the verification screen. A 6-digit verification code is sent to the email address you provided when signing up. The code is valid for 24 hours.

Enter the code you received in the field on the screen and click “Verify”. After successful verification, you will be redirected to the sign-in page.

If the code does not arrive:

Check the “Spam” folder in your mailbox.

Click “Resend code”. After you click it, the button is unavailable for 60 seconds — this protects against repeated sending. When the timer runs out, you can request the email again: while the previous code is still valid, the new email contains the same code.

If several attempts fail, contact your organization’s administrator.

Password recovery

If you have forgotten your password, click the “Forgot password?” link on the sign-in screen.

Step 1. Request a code. Enter your email address and click “Send reset code”. If the address is registered, you will receive an email with a 6-digit recovery code (valid for 24 hours). For security reasons, the screen does not reveal whether the account exists: the confirmation message is the same for any address.

If the email does not arrive:

Check the “Spam” folder.

60 seconds after the request, the “Send reset code” button becomes available again — a repeated request sends a new email; while the previous code is still valid, the email contains the same code, and after three wrong attempts or once the code expires, it contains a new one.

If you see a message about too many requests or about the service being temporarily unavailable, the code was not sent — wait a minute and try again.

Step 2. Enter the code and a new password. Go to the password reset page (the link is in the email, or you can enter the code manually). Fill in:

Email — your account’s email address.

Code — the 6-digit code from the email.

New password — choose a new strong password. The strength indicator helps you assess it.

Confirm password — enter the new password again.

If the code is not accepted:

An invalid code, an expired code, and too many entry attempts all show the same message: you need to request a new code. For security reasons, the exact reason is not given.

Next to the error message, there is a “Request a new code” link — it takes you back to the code request screen.

You can request a new code 60 seconds after the previous request, just as in step 1.

After three wrong attempts, the previous code is no longer accepted — request a new one.

After a successful reset, you will be redirected to the sign-in page, where you can sign in with your new password.

Onboarding

The first time you sign in, you will be offered a short initial setup — a wizard with several steps. It helps Iskra tailor its work to your tasks and level of experience. It takes less than a minute.

The questions, the answer options, and the completed learning profile are shown in the selected interface language. If you change the language while the wizard is open, its content updates automatically.

A progress bar at the top of the screen shows how many steps you have completed and how many remain. There are three short steps in total; on each one, you choose from a set of suggested options.

Wizard steps:

What are your roles? — choose one or more roles from the cloud of options (developer, analyst, project manager, designer, lawyer, teacher, and others). You must select at least one option; there is no upper limit.

Work tasks — choose the tasks you plan to handle with AI: data analysis, writing, coding and automation, presentations, research, planning, translation, reports, and so on. Select at least one option.

AI experience — a single choice: “Never used”, “Tried”, or “Active user”.

Navigating the wizard:

“Next” / “Continue” button — goes to the next step. The button becomes active once at least one option is selected on the current step.

“Back” button — returns to the previous step so you can change your answers. All your choices are kept between steps.

On the last step, the wizard finishes and your answers are saved — Iskra will use them to give more relevant tips and choose examples from your field.

If the wizard is not mandatory, the first step has a “Skip” button, and there is a close (×) button in the top right corner. You can also close the window by pressing Esc or by clicking outside the modal window.

Ways to access Iskra: web, desktop, and mobile app

Iskra is available in the browser — this is the main way to work with it. The desktop app adds local folder synchronization (see the “Folder sync” section). If builds are published on your installation, you can download it via direct links: macOS, Windows, Linux. The /api/desktop/download path without the platform parameter is not a universal direct link: the server guesses the platform from the browser and returns an error if the platform is unknown. If the user’s platform is unknown, give the separate links above or ask which platform they use first. The “IskraBot” mobile app for Android and iOS is available in the official app stores — links and operating modes are listed in the mobile app help. For compatibility with your corporate installation and the address to connect to, check with its administrator.

Product name, model name, and the Anthill label

The official product name is Iskra. The technical domain, the model name in logs or in the interface, and internal labels (for example, Anthill, the name of the agent runtime) are not the product name and do not confirm any history of the brand. An unknown history of the name must not be filled in by guesswork.

The language model actually used depends on the installation, the pricing plan, and the response power mode (see “Automatic power level selection” in the “Chat with the assistant” section):

the platform routes the request between models on its own. The exact list of models for a specific installation is known to its administrator; without such confirmation, a specific model should not be named.

Overview

Chat is the main way to interact with Iskra. Here you ask questions, get answers, attach files, and manage your chat history.

Creating and managing apps

The app management section is available when the Apps module is enabled (APPS_ENABLED). If app creation is turned off, you can still view and manage existing apps in the available section. Creating and publishing from a chat additionally require the Apps runtime to be enabled (APPS_RUNTIME_ENABLED), along with the related skill and scenario mechanisms.

When creation is available, open the “Apps” section and click “Create app”. A new chat opens with the scenario attached and a draft request. Add your idea to the text and send it: the draft is not sent automatically.

The assistant clarifies the task and shows a brief description of the future app. Files are created only after you confirm separately. The assistant then builds the app in an isolated environment without internet access, publishes it from the current chat’s files, and waits for the automatic check. The check opens the main page in a real browser and rejects the version if it does not start, responds with an error, or writes a JavaScript error to the console.

The response contains the app ID, the version number, and the publication status. ready means that the version passed the check successfully; whether it has become current is shown by

ready, but the previous version remained current, for example, while approval is required. probing means that the check is still in progress. While the status is probing, do not ask to publish the same version number again: the accepted version is not lost. If the check fails, the assistant receives safe diagnostics, fixes the sources, and tries to publish the next version, but no more than twice after the first publication. If the app cannot be fixed within these attempts, the assistant honestly reports the error and saves a fresh archive of the sources so that work can continue, without claiming success. Do not put access keys or passwords in the app’s sources or files.

A finished app starts out private and is available only to its owner. The scenario does not make the app public. For public access, open the app’s card in the “Apps” section and confirm the change separately: Iskra shows a red warning and asks you to agree explicitly. The section lists the apps available to the active profile, with their access mode, overall status, current version, and time of the last change. Click a card to open the app page.

To refine the app, it is more convenient to return to the original chat: the author’s sources are kept there. If you continue in a new chat, first download a fresh source archive in the original chat, then attach exactly that ZIP file together with the address of the app card. The assistant checks the archive before unpacking it and making any changes. The published version that the platform runs does not, by itself, allow the author’s sources to be restored.

If the system skill for creating apps is turned off in the active profile, the button does not bypass this setting. Turn the skill on and open the creation scenario again.

App versions and diagnostics

On the app page, the “Versions” section shows the publications that passed the check. The current version is marked separately, and the app owner can activate another ready version, including rolling back to the previous one. A pending version does not serve users yet: its label says whether it will be activated automatically after the check or will require manual activation. If the check fails, a safe description of the reason appears under the version. Fix the app and publish a new version; a failed publication does not become current.

Do not confuse the two status blocks:

App status — whether the app is enabled for users: “Active” or “Stopped”.

Runtime — what happens when the app runs: “Deploying”, “Running”, “Unhealthy”, “Stopped”, or “Failed”. This block also shows the running version, the number of restarts, and the time of the last report.

The app owner and the owner or administrator of its organization can see status details, the latest app messages, and the change history. The history shows when the app was created, when a version was published or activated, and when the app was started, stopped, or had its access changed. Only the app owner sees pending versions together with the reason a check failed. A member who can only view the app does not receive diagnostic messages or the management history.

Opening an app and access modes

The “Open app” button appears only if app sign-in is configured on the platform and the app has a current version and the “Active” status. It opens the app sign-in page in a new tab. If your account has several suitable profiles, Iskra asks you to choose which one to continue in; the recommended profile is marked. After you choose, the app itself opens, and the management page stays in the original tab.

The owner of any app can select the “Owner only” mode right away, and only the owner can initiate a request for the “Everyone” mode. The “Organization” mode is also available, but only if the app belongs to an organization:

Owner only — only the app owner can open the app.

Organization — only for an organization’s app; members of that organization can open it.

Everyone — the public address is available to everyone, including people who are not signed in to Iskra.

An organization Owner or Administrator who does not own the app can switch it to the “Owner only” or “Organization” mode but cannot personally initiate moving someone else’s app to the “Everyone” mode. First, the app owner sends an explicit request, and only then can an Owner or Administrator of the same organization who does not own the app approve the version specified in that request.

This is exactly what the warning “The app will become available to everyone, including unauthenticated users” means: opening the app will not require an account, sign-in, or organization membership. When public mode is requested or approved, Iskra shows a separate red warning. The consent checkbox is initially cleared, and the confirm button is unavailable until you select it. Closing the window or a failed attempt does not save your consent — the next time, you need to select the checkbox again.

The owner of an app in a personal profile turns on public access immediately. The same happens if the owner of an organization’s app also holds the Owner or Administrator role in that organization: in both cases, the app owner’s consent alone is enough, and no separate approver is needed. If an organization’s app is owned by a regular Member, they confirm the warning and send a request, and an organization Owner or Administrator who does not own the app separately confirms the same warning when approving it. Only in this two-step case are two independent consents required. Until approval, the previous mode stays in effect. The pending request is shown with the version number and time; while the app is not yet public, the owner can explicitly click “Cancel public access request”.

Public access approval applies to a specific version. A new version of a public app does not inherit the previous approval: until it is approved separately, users keep getting the previously approved version. The owner of a personal app, or an app owner who holds the organization Owner or Administrator role, activates a new version and approves it themselves in a single action.

For an app owned by a regular Member, the saved request lets an Owner or Administrator of the same organization who does not own the app approve a new ready version of an already public app. Next to each ready non-current version, they see the “Approve version …” action. It opens the red warning again, with the consent checkbox cleared. After confirmation, the selected version becomes current; until then, the previously approved version keeps running.

You can activate or approve a ready version only while the app is running. If the app is stopped, first click “Start”, then repeat the activation or approval. If you try the action before that, Iskra shows a hint: “The app is stopped. Start the app first, then try the action again”.

Stopping, starting, deleting, and notifications

The app owner — and, in an organization, also an Owner or Administrator — can stop the app and start it again. Before stopping it, Iskra warns that the app will be unavailable to users until it is started again. The status in the “Runtime” block may update a little later than the overall status.

Deletion requires a separate confirmation and cannot be undone: the app, its settings, and its active sessions become unavailable, and the app cannot be opened again.

When the runtime first enters the “Unhealthy” or “Failed” state, whoever owns the app at the moment of this transition receives a notification with a link to the app page. Repeated reports of the same failure state do not create duplicates. If the app recovers and then enters a failure state again, Iskra sends a new notification.

Chat interface

After you sign in, you land on the main screen with the message input field. Type your question or request and send it to the assistant.

The input field is at the bottom of the screen. It automatically grows taller as you type multi-line text.

Sending a message:

Enter — send the message.

Shift + Enter — line break (a new line without sending).

After a few exchanges, a light-colored suggestion may appear in the empty field — a likely next message on your behalf. It appears after Iskra finishes its response and takes the current chat into account. If there is no suitable continuation, the field stays empty.

Enter — send the displayed suggestion right away.

Tab, the Right Arrow key, or tapping the suggestion — move it into the field so you can edit it and then send it as usual.

Escape — hide the suggestion. You can also simply start typing your own text: the suggestion disappears and does not replace what you type.

The suggestion appears only when the field is empty and has no attachments. You can turn it off under Settings → App. The send button becomes available once you move the suggestion into the field or type your own text.

While the assistant is generating a response, a loading indicator is shown. You can stop generation by clicking the stop button next to the input field. If you scroll the feed up while Iskra is responding, new parts of the response no longer pull the screen down; to follow the end again, scroll the feed down or click the round arrow button.

In a chat, an interactive task that has started keeps going: Iskra does not ask whether to continue and does not force it to stop just because of total token usage, total time, or the number of steps. This is the default behavior, and it also applies to the parts of the task delegated to subagents. No separate setting is needed.

The monthly usage limit is checked when you send a new request. If it runs out during a task that has already been accepted, Iskra may continue the task, and all further usage is counted. You can stop the work with the stop button. Iskra still asks clarifying questions and requests the necessary confirmations; loop protection and the restrictions of the working folder and of individual operations remain in place.

API calls, scheduled tasks, and apps keep their own time and usage limits.

Only one response can be generated at a time in a single chat. If the same chat is open in another browser tab, that tab automatically connects to the ongoing generation and shows its progress. If you send a new message or start a recap in the second tab while a response is still being generated, Iskra does not create a second parallel response: the current tab connects to the active stream, and the text you typed stays in the input field. When you edit a message that has already been sent, the edited text is not lost either — the editor reopens with what you entered.

Capability strip. While the chat has no messages, a strip of cards labeled “What can Iskra do?” is shown — one card per topic (Memory, files, skills, web search, and so on). On a wide screen, the strip is below the input field; on a narrow one, it is above the greeting. The strip scrolls by itself; the “back” and “forward” arrows are on either side of the cards, and on the line below is a button that pauses and resumes auto-scrolling. Next to that button are page indicator dots, and if there are many pages (for example, on a narrow screen, where the cards are shown one at a time), a counter such as “3 / 21” is shown instead of the dots. Auto-scrolling pauses while the pointer is over the cards or while a card is selected with the keyboard. If your operating system has the setting for reducing animations turned on (sometimes called “Reduce motion”), auto-scrolling is off from the start — you can switch between cards only manually, with the arrows or dots. Clicking a card opens the “What Iskra can do” page: it takes up the whole screen, with a list of topics grouped by section on the left and sample ready-made phrases for the selected topic on the right; you can return to the chat with the close (×) button in the top right corner. The “Try it” button on a sample inserts the phrase into the chat input field but does not send it — you need to send it yourself or add to the text first. If the input field already contained a saved draft, the phrase replaces it.

Voice input and audio. A separate voice input button is not a core feature of every installation. If it is enabled in your version of the interface, the microphone button is on the right in the row below the input field. While recording, the field is “listening”: its outline turns gold and pulses in time with your voice, and the placeholder reads “Iskra is listening”. Stop recording with the microphone button or by clicking anywhere in the input field — the speech is transcribed and the message is sent right away; if the field already contained typed text, it is sent together with the dictated text. If you are still typing a word while transcription is in progress, the voice fragment is added once you finish typing; the requested send is then performed. If you change your mind, click “Cancel” next to the microphone (or press Esc on the keyboard) while recording: the recording is canceled, and nothing is transcribed or sent. If the message cannot be sent right now (for example, an attachment is still uploading or has failed to upload), the transcribed text simply appears in the field. On a phone, an empty chat opens with a large microphone sitting on the top edge of the input field: tap it and Iskra listens (the field shows the status, and a timer appears next to the small microphone button); tap again or touch the field, and the recording is transcribed and sent right away. The row of buttons below the field has “Into the field” (stop and put the text into the field without sending) and “Cancel” (discard the recording), and the small microphone button in that row also stops and sends; the attach file and mode buttons are hidden while recording and return once it stops. In an ongoing chat on a phone, there is no large microphone: the input field “listens” the same way as on a computer, and stopping with the microphone button or by touching the field sends the message right away. A recording lasts no longer than five minutes: when the time is up, it stops by itself, and the text is placed in the field without being sent. Where voice input is enabled, the microphone button is also available in the form for answering Iskra’s clarifying questions: the dictated text is inserted into the field next to the button, does not change the answers to other questions, and does not submit the form — while recording or transcription is in progress, submitting and the answer options are temporarily unavailable, and you need to submit the completed form yourself. If a recording fails to be transcribed, the “Retry transcription” and “Discard recording” buttons appear in place of the microphone button, so you do not need to dictate again. In the main chat field, the recording is kept until it is transcribed successfully, discarded, or the page is reloaded. In a question form, retrying is available only while that question is open: once you submit the answer, the time runs out, or the question is closed in another tab, the recording is deleted. The microphone is also available in the feedback form, which opens from the profile menu (“Help” → “Feedback”), when you rate a response, and with the “Report a problem” button: the dictated text appears in the message field, you can edit it before sending, and the recording itself is not attached to your request. While recording or transcription is in progress, the “Send” button is unavailable; if transcription fails, the same retry and discard buttons appear in place of the microphone, and you can type the text manually. An attachment in this form is optional and does not replace the text: a request without text is not sent. If there is no microphone button, send text or attach supported files as usual.

If the recording contains no speech — you were silent or the microphone captured only silence — Iskra reports “No speech detected”: no text is inserted into the field, no message is sent, and there is no need to retry transcription — just dictate again. For a recording that has sound but no words (noise, music), Iskra also tries to respond with “No speech detected”, but errors are possible in this case.

Other messages state the cause separately. “The microphone produced no audio” means that the browser created an empty recording: check whether another app is using the microphone, and record the phrase again. “The recording has no audio to transcribe” means that the resulting file cannot be read as audio; Iskra does not keep such a recording for a retry. If the transcription service itself is unavailable or the network connection drops, the recording stays at the microphone button, and you can send it for transcription again without dictating anew.

Single draft. The text, attached files, selected skills, scenario, and collected quotes all belong to a single draft. Iskra clears it only after

the send has been accepted; if the message could not be sent or the chat being created could not be opened, the content and the selected chips stay at the input field. If a request is hard to phrase, ask for help with the /prompt skill — Iskra asks a few short clarifying questions and then completes the task right away.

Choosing a skill or scenario. On the new chat screen, and when you create a personal chat inside a project, there is a “/ Skills” button next to the input field. It opens a searchable list of available skills and scenarios: you can find a skill by its human-readable name (including a Russian one), by its technical name, or by the description of its purpose — for example, “draw” finds the image generation skill. The list and the selected chip show the skill’s human-readable name; the technical name stays next to it in small text. You can choose one or more skills and, if needed, one scenario — your selection appears below the input field as chips and applies to the next message. You can remove an unneeded chip with its × button. In a chat that has already started, this button is not available: there, you add a skill with the “Add skill” button in the “Conversation skills” window (opened by the “Skills” button in the bottom row of the request panel), and you attach a scenario to a message from the skill launch window (the ▶ button in the skill row). Searching with / directly in the input field works on all screens: after you make a selection, the /name stays visible next to the chip, and such a call can be sent without any additional text.

You select a skill as a whole — the individual tools inside a skill are not shown in the list, and you do not need to choose them: Iskra decides which of them suits your task as it works.

After sending, the skill call does not disappear: it stays in the message text as /name, and a persistent skill chip is shown under the message. So copying the message also carries over the call itself — if you paste such text into the input field, the skill is attached as a chip automatically. The call is kept when a response is regenerated and when the chat is branched, and the chip is also visible in a chat opened via a read-only link. To remove a skill, edit the message and delete the /name token from its text.

Message statuses. Each of your messages shows a small indicator:

Sent — the platform has accepted the message; the response has either already arrived or is being generated right now.

Sending — the message is being transmitted; the usual actions (fork, hide, rate) are temporarily unavailable.

Error — the message was not delivered, or the assistant could not respond. The message itself is highlighted with a red border, and if details are available, an “Error details” button is shown. A special case is when the monthly token limit is exhausted: the message is marked as an error but is not colored red; an information card is shown instead.

Hidden — the message has been removed from the feed manually; see “Message actions” below.

The limit card offers the next plan up from your current one and states its price: from the free Start plan, this is a subscription to the first paid plan; from a paid plan, it is an upgrade to the next tier. On the “Pro” plan and above, there is a second line under the upgrade offer — “Reset … limit” — which immediately restores your plan’s full monthly token allowance for the plan’s full price, without waiting for the window to renew automatically (for details, see “Monthly limit reset” in the “Settings” section). If you are already on the highest personal plan (Ultra), there is no upgrade, and the reset becomes the card’s only paid action. The email sent when you reach 100% of your monthly limit uses the same upgrade offer logic.

If the list of plans fails to load at that moment, the card cannot name a specific plan and price, but the button that leads to the plans remains — you can choose a suitable plan on the pricing page.

Everything described above applies to the personal workspace. In an organization workspace, the plan card is not shown: there, the limit belongs to the organization, not to you personally, so instead of a personal subscription offer, you see a message about which quota is exhausted — your personal member quota, the organization’s shared pool, or the monthly limit of the organization’s plan. What to do next depends on your role: for the organization owner, the message suggests the available action; for other members, it says whom to contact. For details, see the “Organization” section.

Choosing the power level

In “Auto” mode, Iskra starts a chat at the “Fast” or “Smart” level. If the task becomes more complex, “Fast” may switch to “Smart”. After that, the automatic mode keeps “Smart” until the end of the chat, including for short follow-up questions and topic changes. Only you can turn on the maximum “Very smart” level.

You can also select a lower level manually. Switching back to “Auto” restores the automatic level already reached: for example, after a manual “Fast”, “Smart” turns on again if Iskra had already chosen it earlier. Reloading the page or clearing the context does not reset this choice.

Only you can turn on increased power: select the “Very smart” level with the “Complexity” button in the bottom row of the request panel (see below). This level charges tokens at an increased rate, and its price is stated on a separate line right in the expanded row of levels — “Very smart — manual only, ×N charge”. So you can see it before you choose the level, not only after you are charged. The selected level applies from the next message, belongs to the current chat, is restored when you reopen the chat, and stays in effect until you change it yourself.

After a response at the maximum level, the “Complexity” button shows a “×N” mark. The number N is the actual usage rate of that response; expand the complexity control to read the explanation. If you change the level or the LLM profile, the mark disappears immediately: the previous value no longer promises anything, and the newly selected option states the price of the next response itself.

If the maximum level is temporarily unavailable, Iskra continues the response on a regular model. The mark changes according to the rate of the model that actually performed the subsequent steps. Steps that were already completed are counted according to the mode in which they were actually performed.

On some platforms, our own server handles the “Fast” and “Smart” levels: while the server is free, the response runs on it, and each level shows the price of the model that will handle it. If the server is busy at the moment or the chat contains images that its model cannot recognize, another model handles the request. Availability is checked before each next step of the task: if our own server has become free, the work can return to it. The response already received and the results of actions are kept. Usage for each step is calculated at the rate of the model that actually performed it; if the rates are equal, the switch does not change the cost.

The explanatory line above the response parameters does not always appear. Iskra always mentions images. It mentions a busy server only if you pinned the “Fast” or “Smart” level yourself: in that case, the promise was not kept, and this has to be stated. If the level was chosen automatically, there is no separate line about a busy server. Usage is still determined by the model that actually performed each step.

Iskra can also reason more deeply without turning on increased power. A “think” block by itself does not mean an increased rate — go by the “×N” mark on the “Complexity” button.

If Iskra delegates work to a subagent, a reasoning model is selected for it based on the platform settings and the selected LLM profile. It may differ from the model used for the main response. The assignment scope describes the subagent’s work, while its actual level and cost are shown in its card.

The subagent card shows the assignment scope, the current level, and a switch to a fallback, if one was needed. This information is kept in the history, the shared link, and the downloaded chat. A subagent’s switch to a fallback changes its card; the main panel keeps showing the level of the main response. Usage is counted by the model that actually performed each step. When usage is calculated, the model rate shown is multiplied by the rate of the selected LLM profile, if one applies.

Attaching files

You can attach files to a message so that the assistant can analyze their content.

Click the “+” button to the left of the input field. A menu opens where you choose the file type:

Document — text files and documents.

Image — image files (available when you use a model that supports images).

You can also drag a file anywhere into the chat area or paste an image from the clipboard (Ctrl+V / Cmd+V).

Supported document formats:

Text and data: TXT, MD/MARKDOWN, CSV/TSV, JSON, XML, YAML/YML, HTML/HTM

Office (Writer): DOC, DOCX, DOCM, DOT, DOTX, ODT, FODT, OTT, RTF

Office (Calc): XLS, XLSX, XLSM, XLT, XLTX, ODS, FODS, OTS

Office (Impress): PPT, PPTX, PPTM, PPS, PPSX, POT, POTX, ODP, FODP, OTP

Office (Draw): ODG, FODG, OTG

PDF, DJVU/DJV, EPUB, FB2, and ZIP

Iskra can try to parse other formats with agent tools — the result depends on the format and the task.

When reading a PDF, Iskra checks whether the embedded text layer can be trusted. If the layer is damaged and contains unreadable characters instead of text, Iskra does not use them in the response and instead recognizes the page from its image. Recognition may continue in the background: ask Iskra to read the file again in a few seconds. If a page could not be recognized, the assistant must say directly that the page needs visual verification, rather than draw conclusions from the damaged text.

Supported image formats:

JPEG, PNG, WebP, AVIF, GIF, BMP, TIFF, HEIC

If the browser does not support an image format, Iskra prepares a compatible copy for the preview. The original file remains available for download.

Before sending, attached files are shown above the input field as cards. Each card shows the file type, name, and size; you can remove a file by clicking the close icon.

Attachment preview

Click a file card in a sent message to open the preview. PDF, DJVU, Word/Office documents, presentations, spreadsheets, Markdown, images, and code are displayed right in the preview window. A link to one of the chat’s Markdown or PDF files in the response text also opens this window rather than a new tab; the browser’s “Back” button closes it and returns you to the same chat. Links to files in other formats open in a new tab, as before. Iskra shows DJVU, supported documents, and

presentations as a PDF version generated on the server. If conversion or PDF viewing is unavailable, the interface shows diagnostics and offers to download the original.

In the PDF preview, individual pages — and, for presentations, slides — are shown on a contrasting background, with visible borders and gaps between them.

Spreadsheets open not as a PDF but in a separate read-only mode. It supports XLSX, XLS, XLSM, Excel templates, ODS/FODS/OTS, CSV, and TSV. The viewer supports sheets, selecting and copying cells, merged cells, row and column sizes and hidden rows and columns, frozen panes, formatting, links, and notes. When you select a calculated cell, its source formula is shown above the table, while the cell itself keeps the value calculated by LibreOffice. You can view drop-down lists and checkboxes without changing the document. Charts and images are shown as static PNGs.

Editing, interactive charts, pivot tables, and sparklines are not supported in this mode. If the file has more than 200,000 non-empty cells, Iskra shows the available part along with a persistent note about the limit. If the spreadsheet preview could not be built in full, you can still download the original file; the extracted text remains available where it was created during file processing.

In the preview window and on file cards, the download button may open a menu. The Original option downloads the uploaded file unchanged. For DJVU, supported documents, and presentations, a PDF option is also available — the same version that is used for the preview. For spreadsheet files, no PDF option is offered. In the Iskra desktop app, the selected file is saved to the system “Downloads” folder, and a notification with its name appears at the bottom of the screen; if saving fails, an error message is shown instead. The chat stays open throughout.

Legacy IPF presentations

.ipf is an outdated, opaque presentation format. Iskra does not open or edit it directly: instead of a preview, it shows an explanation and a button to download the original. For an active IPF file among the current chat’s files, an explicit “Create PPTX” action is always available there as well. For conversion, the source IPF file must not exceed 10 MiB.

The conversion creates a permanent sibling file with the same name and the .pptx extension: for example, plan.ipf becomes plan.pptx. The source IPF file, the old message, and the link in it do not change, and the new PPTX is not added to the old message. Iskra never overwrites an existing PPTX; in case of a conflict, the interface offers to open the previously created file. If that file is in the Trash, a “Restore PPTX” button appears — after it is restored, the file opens in the regular preview. The new PPTX appears in the chat working folder and stays there as a regular file.

When you select a file, Iskra immediately checks the file size and the available storage space, but the file itself is uploaded only after you send the message. During a batch upload, each card shows its own status: “Waiting to upload”, “Uploading”, “Uploaded”, or “Upload failed”; for a file that is being uploaded, the progress is shown as well. The send button is temporarily blocked. A card with an error shows the full reason. If the failure is temporary, a “Retry upload” button is available; if the file is too large, you need to remove it and attach a smaller version. The other files in the batch remain visible. If you closed the chat or removed the card before sending, the file does not take up storage space.

Once a message with an attachment is accepted, the file immediately appears in the “Chat files” area, in the chat’s working folder — you do not need to wait for the response to finish. It stays visible after you refresh the page.

After you send a message with images, Iskra may first show the progress of preparing the attachments — how many files have already been processed and the name of the file that was just completed. The assistant’s response starts once this preparation is finished.

File limits

The exact size limit for a single file and the total storage space depend on the active profile’s plan, the organization package, and the technical settings of the specific installation. The current values are shown in the interface: in the subscription settings, on the pricing page, and in the error dialogs during upload.

The platform administrator can also set a technical upper limit on file size for the entire installation. So the actual limit for a single file may be lower than the value stated for your plan.

If the selected file exceeds the available limit or the profile has run out of file storage, Iskra shows a dialog with the current restriction. If the restriction comes from the plan, the dialog offers a link to the plans. The check works the same way whether you select a file through the menu, drag and drop it, paste it from the clipboard, or add a screenshot. Right before sending, Iskra repeats the check to account for files that may have been uploaded in parallel in other chats.

You do not need to split a large document into parts in advance: Iskra works with its content in parts and, where this is enabled on the installation, delegates reading to subagents when needed, instead of loading the whole file into the context (see “Reading large materials” in the “Agent mode” section).

Office documents in Word, Excel, and PowerPoint formats, as well as the supported OpenDocument formats, are subject to an additional processing limit of 50 MiB (shown as 50 MB in the interface). It is checked when you select the file, before its content is transferred, even if your plan allows a larger size. CSV and TSV files are not subject to this limit. The rejection dialog shows the file name, its size, and the allowed limit. Compress or split the file; if you need a document or presentation to be read, you can export it to PDF. Upgrading your plan does not raise this limit, and uploading the original file again will not help.

Files that have already been saved are also checked before a message is sent. If a file passes the size check but its useful content is too large to process, reduce the document or split it.

If upgrading to a higher plan solves the problem, the dialog shows a comparison of your current plan and the recommended plan: file storage, file size, MT limits, and available features. The upgrade button leads directly to payment for the suitable plan.

Request panel

To the right of the conversation is the request panel. It shows exactly what will handle your next request and lets you step in. By default, Iskra decides everything on its own, so you do not need to open the panel just to chat.

The panel consists of two areas that are always visible and a fixed bottom section:

Connected folders: what Iskra sees. This is the project card (if the conversation is open in a project) and the connected Memory folders with their access mode. When there are more folders than fit, the “Show all” row opens a window with the full list.

Chat files: the tree of the conversation’s working folder, with message attachments and files created by the agent. The area header has the “Starred” and “Trash” icons (they appear when there is something to show) and, when the folder has files, the Memory icon “Go to Memory”.

Footer: two fixed rows. The top row holds the profile and Antonym (it appears when the organization has allowed profiles or private mode is available). The bottom row has, on the left, the conversation history fill ring (which is also the compact button), “Clear”, “Skills” and “Pinned notes”, and on the right, the response parameters: complexity and specialist.

Conversation skills and pinned notes are not listed in the panel column: they are managed in two windows opened by the buttons in the bottom row (see “Footer: context and response parameters” below). Each of these buttons has a number in its corner: how many skills are connected to the next request and how many notes are in effect in the conversation. When there are none, no number is shown, except when Iskra itself has changed the set: then the yellow mark is shown even at zero.

The “Chat files” area takes up all the height left over by “Connected folders”, and the file list scrolls inside it, so a dozen conversation files do not stretch the panel across the whole screen. On a short screen, where the list does not have room for even a few rows, the entire panel column scrolls instead. The footer with the response parameters is fixed and does not scroll.

Automatic and manual mode

The skill set mode is shown in the “Conversation skills” window: under the title there is a mode badge and a hint line.

“auto mode”: Iskra picks the set; the hint reads “Skills are picked automatically as you chat”.

“manual mode”: you have defined the set, and automation no longer changes it; the hint reads “Automation no longer changes this set”. Next to the badge is the “Back to automatic” link.

Manual mode is visible even without opening the window: the “Skills” button in the panel’s bottom row has a yellow outline. While Iskra is picking the set, the button has no outline.

The set switches to manual mode on any intervention from you: adding a skill with the “Add a skill” button or removing one with the cross. The very first such action locks the entire displayed set to the conversation, including your edit; you do not need a separate button for this.

The profile, complexity level and Antonym setting in the footer are pinned as soon as you choose them and apply to this conversation. Choosing one or two specialists first changes a draft and takes effect when you click “Save”. To hand the decision back to Iskra, choose “Auto” in the complexity picker or the specialist editor; for the profile, choose the organization row marked “set by the organization”.

Specialist: a response role, not a skill

A specialist sets the role of the response: for example, lawyer, analyst or developer. You can choose up to two specialists: the primary one sets the working approach and how the answer is presented, and the secondary one adds subject-matter expertise. It is still one task and one response, not two independent performers. Specialist names and descriptions follow the interface language. Changing the language keeps the selected roles and their order. A specialist is not a skill and does not add tools: it is a selectable response role, whereas a skill is a separate add-on capability with its own tools and settings. You can use them together and configure them independently.

You do not choose which tool of a connected skill to use: Iskra picks the tool itself based on the task. Your choice is which skills are in the conversation’s set (see “The ‘Conversation skills’ window”), not which of their tools to call.

In “Auto” mode, Iskra chooses one or two specialists at the start of the conversation. This lineup and its order are kept for subsequent messages, even if the topic changes. “General” also counts as a completed choice; you can assign a different role manually later. The letter A on the icon means automatic selection; click the icon to see the selected roles and their descriptions. If the selection fails, General answers, and the expanded row explains why and suggests choosing a specialist yourself.

In the expanded row, check one or two specialists and click “Save”. The first one you check becomes the primary; you can change the order and remove an extra role. “General” replaces the whole set rather than being added to another role. Until you save, the checkboxes change only the draft; closing the row discards unsaved changes. The pinned choice applies to subsequent messages until you switch back to “Auto”. Switching back restores the original automatic lineup. If you sent the first message with a manually selected specialist, General answers after you switch back to “Auto”.

If the setting was changed in another tab at the same time, the panel shows the server version and your draft. Load the current version, or confirm your choice again and click “Save”: the conflict by itself does not overwrite anything.

A pair is shown as two overlapping icons; the tooltip shows the order of the roles. The specialist is chosen only in the request panel footer; there is no separate button next to the message field. A second specialist adds instruction text and may increase input tokens; there is no separate charge “for the second specialist”.

Independent conversation settings

The “Connected folders” area has no mode badge: connected Memory folders are always your choice, automation does not revise them, and there is nothing to hand back here.

Manual skill set mode, the specialist and the pinned response parameters are independent of each other: you can define the skill set, pin a specialist and leave the model choice to automation. All settings apply only to the current conversation, persist across page reloads and are removed only by an explicit action on your part; they never revert to “auto” on their own. A new conversation starts in automatic mode.

You can also set them before sending the first message: open the panel on the start screen, and your choices will apply to the conversation being created.

When the panel is visible

On a wide screen (1280 pixels or more) with a mouse, the panel is open in the conversation right away; you do not need to call it up. If you close it, it stays closed in other conversations too: this is one decision for the whole app, and your browser remembers it until you open the panel again. On narrow screens and touch devices, the panel slides out over the conversation, so it does not open by itself: open it with the arrow at the right edge of the screen or the panel button in the top bar, and close it with the cross or the “Back to chat” link.

While the panel is collapsed, a summary line above the input field shows what is hidden in it: “In this chat: 3 files · 5 skills · context 58% free”. Clicking the line also opens the panel, after which the line disappears, since everything is now visible in the panel itself. The summary appears only when there is something to report: in an empty conversation with no files or selected skills, it is not shown.

During a reply

The response parameters in the footer (profile, complexity, specialist and Antonym) are available during a reply too. A change applies to the next reply. A reply in progress

continues the way it started: switching does not interrupt or redo it.

The other settings are locked during this time: connecting skills, Memory folders and files is dimmed with the hint “Wait for the reply to finish”. Viewing files and scrolling work as usual: files appear in the chat working folder precisely while the work is in progress, and that is exactly when you need to look at them.

If the same conversation is open in another tab and the settings were changed there, the panel shows the message “The settings were changed in another tab” and redraws with the fresh values.

Some restrictions are permanent: the conversation is archived or read-only, the organization’s subscription has ended, or the message limit is exhausted. In that case, the controls are unavailable right away, without waiting for a reply, and the reason (“The conversation is archived — settings cannot be changed”, “The message limit is exhausted” and the like) appears as a tooltip on the “+” button. In an archived conversation, and in a conversation from an archived project, connected Memory is shown read-only: the full list of folders is visible, but the rows have no crosses or “read/write” switches, because the lineup cannot be changed until the conversation is restored anyway. The round response parameter buttons still look normal: they do not write anything, they only expand a row; their tooltip names the reason, and in the expanded row it appears as a line under the switch.

The “Connected folders” area

This area lists the Memory folders that Iskra searches in this conversation. Each connection is shown as a row: a folder icon, the name, a badge with the agent’s access mode (“Read” or “Write”) and a cross that disconnects the source; the cross removes only the connection, while the folder itself and its files stay in Memory. The “+” button in the area header opens a picker with the Memory folders available to you. The full name (and the path, if the folder is nested) appears as a tooltip on hover. You can always narrow the mode to “Read”, but you can widen it only when your permissions and the resource settings allow it (for details, see the “Memory” section); an unavailable connection is marked “unavailable”. While nothing is connected, the area shows the hint “No connections yet”.

If the conversation is open in a project, the first row of the area is the project card: the name, the caption “You own it” or the owner’s name, the number of project folders, and file storage usage as “used / limit”. Folders inherited from the project are not listed in the area: they are managed on the project page. Clicking the card opens the “Connected folders” window on the “Project” tab, which shows the project folders (view only) and “Project files”; the assistant can only read them. The second tab of the window, “Folders and collections”, is the full list of the conversation’s own connections with the same modes and crosses; the “Connect more” button at the bottom opens the folder picker. When there are more connections than fit in the area, the “Show all” row opens the same window.

You can also connect a folder or a single file right while typing: type “@” in the input field and start typing the name, and a list of matching folders and files appears above the field. The selected connection immediately appears as a row in this area, and on the start screen the panel itself opens along with it: the list of connections is no longer shown under the input field, it lives here now.

In the “Connected folders” window, on the “Project” tab, next to the project files heading, you see file storage usage as “used / limit” and a circular indicator; clicking the indicator opens the plan upgrade window.

The “Chat files” area

The “Chat files” area shows the files of the folder the chat works in. It takes up all the height left in the column below “Connected folders”, whether there are many files or none at all, so that the panel does not jump when the first file appears; a long list scrolls inside the area. The Memory icon in the area header (tooltip “Go to Memory”) opens the same folder in the “Memory” section; while the folder has no files, the icon is not shown, since there is nothing to open, and the “Memory” section itself is always available from the left menu. If an existing Memory folder has been chosen for the chat (see “Chat working folder” below), its name appears as a line above the tree; in any case, the working folder is never linked to folders synced from your computer.

Chat working folder

By default, Iskra creates a system-managed folder for a new chat’s files. In a new personal chat, you can choose an existing Memory folder instead: on the start screen or in the request panel, while the chat has no messages or files of its own, click “Choose a folder for files” in the “Chat files” area, and in the window that opens, click “Choose” next to the folder you want.

Until the first message, or the first file uploaded or created by this chat, you can cancel the choice: click the “Cancel workspace selection” cross next to the folder name. After you cancel, folder selection is available again; if you choose nothing, the chat uses the system-managed folder. Connected context folders are managed separately and are kept when you cancel.

One Memory folder can be the working folder for several chats. Files that were already in it or were created by other chats do not prevent you from cancelling the choice in an empty chat. Cancelling keeps the folder itself and all its files where they were.

If the working folder itself has been moved to the trash, its link to the chat is kept. To work with it again or to cancel the choice in an empty chat, first restore the folder in Memory.

After the first message or the chat’s own first file, the working folder is locked to the chat. Deleting messages or moving files to the trash does not bring back the option to choose. To use a different working folder, start a new chat. Deleting the chat does not affect the selected folder itself: the chat disappears, but the folder’s files remain in the “Memory” section.

Result files that the assistant creates while working are shown in the “Chat files” area as a folder tree. You can expand and collapse nested folders individually. Files created by the assistant or a skill have a source icon next to the name (the tooltip says exactly who created the file); your own attachments have no icon, since they are obviously yours. A colored dot before the icon marks files added or changed in this work session.

Actions are collected in the “⋯” menu on the file row (on devices with a mouse, the button appears on hover; on touch devices, it is always visible): download, copy the contents to the clipboard, view version history, delete. If a PDF preview version is available for a non-spreadsheet file, the “Download” item has a submenu where you choose the original or the PDF. Deleted files go to the “Trash” behind the icon in the area header: in its list, each file has “Download” and “Restore”. Messages you have starred are collected behind the “Starred” icon next to it: clicking a row takes you to the message, and the cross removes it from starred messages.

Right below the working files, a “Drafts” section appears, collapsed by default. Iskra keeps intermediate files there, separate from finished results: for example, calculations, the source of a future report, or HTML that is still being checked. While there are no drafts, the panel has no such section either. This is not a loading failure or a lack of permissions: the section appears by itself as soon as Iskra puts the first file in it. If the drafts failed to load, however, the section does appear, with a warning sign in its row and a

“Retry” button inside: otherwise a failure would be indistinguishable from an empty section, and there would be no way to try again. The section expands with a caret, like a regular folder in the tree; the files and folders inside look and work the same way as in the working folder: you can view, download, copy contents, view version history and delete.

Each such file is kept for 30 days from the last change to its contents. A new content version starts a new 30-day period, while viewing, downloading, renaming, moving, deleting and restoring do not extend it. A deleted draft file goes to the shared “Trash” only for the remainder of that period; after it is restored, it appears in “Drafts” again. There are no extra 30 days after deletion.

“Drafts” remain the conversation’s private workspace: they are not included in a public link or the chat’s ZIP export. This section is not related to the message draft in the input field, which is described separately below.

For recurring work, how to choose between skill materials, a Memory folder, Drafts and finished files is covered in the skills guide.

If the name of a new file matches a file in the “Trash”, Iskra automatically saves the new file with a number: for example, report.docx becomes report (2).docx. As a result, Iskra and the agent always show the actual name.

If the conflict arises when changing an existing file, the file is not changed: restore or rename the file with the same name in the “Trash”, or save the result under a different name.

File viewer

Click a file in the “Chat files” tree to open its preview in a separate window. The window displays images, PDFs, Word/Office documents, presentations, spreadsheets in a read-only Excel-like view, Markdown, web pages (HTML), code with syntax highlighting, and audio and video with a built-in player. For supported documents and presentations, the preview opens a PDF version; spreadsheets are never converted to PDF. If the file cannot be previewed at all, the window has a download button.

The preview window opens full screen, edge to edge: the document itself takes up the whole area. The controls are gathered in a narrow bar at the top: the file name, the view mode switch, window size, version history, download and close. The size icon shrinks the view into a window with margins so that the chat is visible behind it, and expands it back to full screen; your choice is remembered and applied to the next files. In windowed mode, you can adjust the size: drag the handle in the bottom-right corner, and double-click it to restore the default size. On a phone, the switch works too, but there is no corner resizing: the window takes up almost the whole screen. The version history window gets the same size. For Markdown and web pages (HTML), the same bar has a “Preview / Source” switch with two icons (an eye and </>): “Preview” shows the formatted result (for HTML, the rendered page), and “Source” shows the source text with highlighting and line numbers.

A Markdown file without an extension also opens in preview if its type is set to Markdown. Plain text with the “text file” type stays in source mode.

File version history

Every save of a working folder file creates a new version, and the previous ones are kept. You can open them in two ways: open the “⋯” menu on the file row in the “Chat files” area and choose “Version history”, or click the same button in the top bar of the

preview window. The button is available for files in the chat working folder and on Drive, wherever you open the file from, and even when there is only one version so far.

The left side of the window lists the versions, newest first: number, date and time, size. The current version has a “current” badge, and a version written during a conversation is marked “from chat”. The selected version opens on the right, read-only: an old version cannot be edited or accidentally saved over the current file. Up to fifty of the most recent versions are shown at once; the “Show more” button loads the rest. If the file has been saved only once, the note “This file has only one version so far” appears under the list.

The “Make current” button brings back the contents of the selected version. Before that, Iskra asks for confirmation and explains what will happen: the contents of the selected version will become the new current version, and the history will be kept. A rollback erases nothing: the restored contents become the newest version, and all the previous ones, including the one that was current before the rollback, stay in the list. So you can always undo a rollback.

If the file was changed while you were viewing the history (for example, Iskra was adding to it at that very moment), the restore does not go through, and you see the message “The file changed while you were viewing history. The list was refreshed — check and retry.” Review the updated list and try again if you still need the selected version.

For very old versions of files that have been in Iskra for a long time, the contents may no longer be stored. The row for such a version stays in the list, but instead of a preview it shows the note “The content of this version is no longer available”, and an attempt to make it current fails; choose another version. For files created now, all versions are kept.

Only active working folder files have a history: a file moved to the “Trash” has no history button, so first restore the file from the trash. Version history is visible to the folder owner (for project files, to the project owner); for other conversation participants, the window reports that the history could not be loaded. Version history is not shown in the “Memory” section or via a public link to the conversation.

The “Boards of this chat” area

Below the chat files is the “Boards of this chat” block: the boards this conversation is linked to, with the board name, its audience, and tabs with their placement status. The agent sees the same list when it works in this conversation, so you can ask it to refine material that is on a board. The block shows only the boards available to you; a board with an external audience (“Named recipients” or “Anyone with the link”) is visible in it to the board’s owner, while others open it by invitation or link, not from the conversation’s inventory. The block does not grant access to a board and does not manage its contents.

You can place a chat file on a board from the “⋯” menu on the file row with the “Publish on a board” item; a new board created this way is linked to this conversation automatically, while placing a file on an existing board does not create a link. A link is also created by your choice in the “Refine in chat” window: this way, the board material returns to work together with its folder, connected to the conversation context. The full rule is described in the “Boards” section.

What boards are, what audiences they can have, how “Follow the file” differs from a pinned version, and what readers get is covered in the “Boards” section.

The “Conversation skills” window

The skill set for the next request is shown in the “Conversation skills” window, which opens with the “Skills” button in the panel’s bottom row. The number in the corner of the button shows how many skills

are connected; skills required by your organization are not included in this number, since they are always in the set. More than nine is shown as “9+”; the exact number is in the button’s tooltip. If Iskra itself changed the set in the last reply, the mark is yellow until you open the window. The button is also available on the start screen, where there is no conversation yet, and in a scheduled conversation; while Iskra is replying, the window still opens, so you can always view the set, but you cannot change it at that time: the window’s buttons are dimmed and state the reason.

Before a message is sent, the window shows the starting skill set for the next request. In automatic mode, Iskra can additionally find and connect one clearly matching installed skill during the reply; the set then updates right during the reply. This connection alone does not lock the skill in for subsequent messages. In manual mode, the set does not expand: Iskra uses only the skills shown.

The set is shown in two groups. At the top is “Connected” with a counter: this is the set you manage, and it is always expanded. Below it is “System” (skills required by your organization) with its own counter; this group is collapsed and expands when you click its row, while Esc closes the window itself. Each row has the skill icon, its name and, next to it (under the name on a phone), a status label or the skill’s slash command, and on the right three controls in a column: ▶ run, ⚙ settings, and ✕ for skills that can be removed; where a control does not apply, its place stays empty so that the column stays aligned. The groups are styled as a table: one border per group, with thin lines between rows. All controls are also keyboard accessible: Tab reaches them, and Enter activates them. While the set is empty, the “Connected” group shows the hint “No skills yet”, and if the organization has enabled system skills, “Nothing connected yet”: the conversation does have skills, you just have not added any yourself.

The ▶ button appears next to a skill that has something to run. For a skill with scenarios, it opens the run window, which lists the manual run (if there is one) and all of the skill’s scenarios. Each scenario has two buttons:

“Run” sends the request right away, and the reply starts in this same conversation;

“Attach” applies the scenario to the message you write next. An attached scenario is shown as a chip under the input field; the same button in the window detaches it. You can attach on the start screen, before the first message, and while Iskra is replying: that is exactly how you prepare the next request. Attaching is unavailable only where there will be no next message at all: in an archived conversation, with read-only access, and when the limit is exhausted.

For a skill without scenarios, ▶ runs it right away, without a window. On the start screen, running is dimmed with the hint “Send a message to the chat first”; the window still opens and shows the reason.

The cross in a row removes the skill from the set. This is the only way to remove a skill, and it is always visible; skills required by the organization and skills that Iskra added itself for the current request have no cross. The gear opens the skill settings; for a skill that is not set up, the gear stays in the same place but is colored yellow, and the reason is written next to the name (under it on a phone), for example, “Requires secrets setup before enabling” or “Awaiting administrator setup”. There is no separate warning icon in the row: ⚠ and ⚙ open the same form.

“Add a skill” at the bottom of the window opens search; this is the only way into it. The search window opens tall, up to the screen height, and does not change size while you type a query: the list scrolls inside. At the top is a search bar for available skills, and below it is the list, with system skills first; each row has the skill icon, its name and slash command, and on the right a symbol instead of a word: “+” in a frame means the skill can be added (“Add to chat”), a green check mark means the skill is already connected (if it can be removed, the check mark turns into a cross under the cursor, and clicking the row removes the skill from the set),

and a lock means a system skill enabled by the organization (clicking does nothing). The row itself is still the button: clicking it adds the skill to the set or removes it, and the search stays open. The “Close” button closes the “Conversation skills” window itself.

Switching to manual mode. There is no separate button for this: your very first action locks the set to the conversation, whether you add a skill with the “Add a skill” button or remove one with the cross. From then on, exactly the displayed set goes into the request, and Iskra no longer revises it; the badge under the window title changes to “manual mode”, and the “Skills” button in the panel’s bottom row gets a yellow outline (see “Automatic and manual mode” above). After that, you can remove an individual skill from the set with the cross, or hand the whole decision back to Iskra with the “Back to automatic” link next to the badge.

Iskra must use a manually defined set: it is not allowed to replace it with its own choice, and if none of the displayed skills fits your request, it must say so directly rather than stay silent or take another skill at its own discretion.

Every change to the set is confirmed by a message in the corner of the screen, such as “Skill … added to chat” or “The … skill is removed from the set”: the list alone does not always make it clear what exactly happened.

The label next to the name (under it on a phone) says why the row cannot be changed. “Added by this request — cannot be removed” means that Iskra itself added the skill for the current request (a launched scenario, a skill named in the message, a knowledge map exploration), and it applies only to this request; if you need it later too, add it separately with the “Add a skill” button. “Required by the organization” means that your organization has enabled the skill. A skill that is not set up shows a yellow setup hint in this place, and other skills show the skill’s slash command of the form /name, which you can use to call the skill right from the input field.

Skills required by your organization are grouped under “System” below “Connected”: they have no cross and cannot be removed, and in skill search they are marked with a lock. If one of them is not set up, the ⚠ icon appears on the “System” row itself, so the collapsed group does not hide the problem.

The conversation’s set is also visible at the input field. If you defined the set manually, the same skills appear as separate chips next to the input field under the “Conversation skills” caption: the same thing in two places. You can connect or remove a skill in either place, and the result is immediately visible in both. The cross on such a chip removes the skill exactly as in the “Conversation skills” window. Skills required by your organization are not shown among these chips, since they are always in the set anyway, and in automatic mode there are no set chips at all: since you have not chosen the lineup yourself yet, showing it as your choice would be inaccurate.

If you ask Iskra “which skills can you see” or something similar, the answer may not match this window, and this is not an error. Such a question is answered not by the conversation set but by the system skill for managing skills: it lists all the skills installed in your profile, including those that are not in the current conversation. The exact lineup for this particular request is shown in the “Conversation skills” window, not in Iskra’s answer to such a question.

While the skill list is still loading, gray placeholder bars stand in for the lineup: skill names arrive in a separate request, and until it responds, the lineup is not shown at all, so that a technical identifier does not flash in place of a name. Memory connections and working folder files behave the same way: until the panel gets a response, it shows placeholders, not “empty” and not an error message. Until the panel has received the lineup of the set, the “Skills” button has no number.

If skill selection did not work or a skill has disappeared, the window shows this as a separate line, for example, “The skill is no longer available”. The line “The skill list is unavailable right now”

means a temporary failure on the platform side: the request will go without skills, but your set is saved and will come back by itself, so you do not need to change it.

The line “Skills will be selected on the next message” is not an error and is therefore shown in a calm color: skill selection has not yet run in this conversation, so for now the set contains only always-on and required skills. You do not need to pick skills manually because of this line; just send a message.

The “Selecting skills” step in the reply feed can appear on more than just the first message: while the conversation still has skills that did not make it into the automatic set, Iskra revises the selection for new text messages. This way, a skill for a changed topic can connect to this same reply, while skills already selected for the set do not drop out. If Iskra used one of the connected skills in the previous reply, the revision is skipped: the selection has just proved itself in practice, and the step does not appear in the feed. If all your skills fit into the set the first time, the step does not repeat.

A skill that disappeared from your list after you assembled the set stays as a row in the lineup and does not block edits to the others: you can remove it with the cross or leave it, since it is not sent with the request anyway.

If a called skill contains additional files with rules or templates, Iskra can use them in later messages of the same conversation too. Before each reply, it connects the current permitted version of the skill: after an update, the new version is used, and after the skill is disabled or access to it is revoked, its files are no longer connected. These internal files are read-only and do not appear in the working folder or in “Drafts”.

The “Pinned notes” window

A pinned note is a short instruction that applies until the end of the conversation: “write in a formal tone”, “the deadline is August 30”, “approved facts are only in 01_Approved_facts.md”. Iskra receives these instructions with every request, so you do not need to repeat them in new messages.

All of the conversation’s pinned notes are collected in the “Pinned notes” window. You can open it in two ways: with the pin button in the bottom row of the request panel (the number in its corner shows how many notes are currently in effect; when there are none, there is no number) or by sending the /pin command without text from the input field. The full text of each note is visible in the window, you can select and copy it, and the cross in a row removes the note. At the bottom of the window is a multi-line input field (long text wraps so that it is easy to reread; the note itself stays a single line: no line break is inserted into it, and pasted line breaks and tabs turn into spaces). Type the text and click the “Add” button or press Enter, including together with Shift or Ctrl; the field keeps focus, so you can add several notes in a row. While nothing is pinned, the hint “No pinned notes yet” appears under the title, and the input field is in place. When there are notes, a reminder about the /pin text and /pin commands and a “Close” button appear under the field; an empty window has neither, and you close it with the cross.

To the left of the button is a microphone (it is there if voice input is enabled on the server): click it, dictate the note and click again; the transcribed text appears in the field, so check it and add it with the button or Enter. If you click “Add” while recording, the window waits for the transcription and adds the text itself. If transcription fails, buttons to retry and to delete the recording appear in place of the microphone, as in the chat field. While Iskra is replying, and also in an archived conversation, recording does not start, but you can always stop a recording in progress. While recording or transcription is in progress, Esc and clicking the dimmed background do not close the window, and Esc during recording cancels the recording, as in the chat field. The cross always closes the window: a recording in progress is cancelled, and if

transcription is in progress, it waits until the next time you open this window in this conversation: the text appears in the field, or the retry and delete buttons appear in place of the microphone.

You can also pin notes without the window: with the /pin the deadline is August 30 command from the input field, and with the “Pin” action in the toolbar that appears when you select a short text fragment in a message (see “Actions on selected text and quotes” below).

A note is meant for a couple of lines, not a document: if the text is too long, Iskra asks you to shorten it, and when the total notes budget is used up, it asks you to remove some. For large materials, there is Memory and the chat working folder; notes do not replace them.

If a note conflicts with the instructions of your profile, project or a connected skill, Iskra follows the note: it is the narrowest instruction, because you set it for this conversation.

Notes survive both “Compact” and “Clear”: they describe the rules of the conversation, not its history, so they go away only when you remove them yourself. Notes belong to the conversation itself and travel with it: branching inherits them along with the messages, moving the conversation to an organization or a project keeps them, and they remain visible and removable in the window. Notes are not included in the conversation export or in a conversation opened via a public link.

While Iskra is replying, and also in an archived conversation, with read-only access and when the limit is exhausted, the window opens, but you cannot pin or remove a note: the button and crosses are dimmed and state the reason, and the microphone does not start recording.

Notes are available in a personal conversation, including a personal project chat. On the start screen, while there is no conversation yet, there is no notes button either: a note belongs to a conversation, and there is nothing to pin it to yet (the “Skills” button is there, because the set applies to the future first request). In a conversation shared with other participants and in temporary conversations, there are no notes at all: the panel has no button, and the /pin command replies that context management is unavailable. Scheduled tasks have no context operations, so /pin is not a command there: the text is sent as a regular message.

Footer: context and response parameters

The panel footer has two rows. The top row holds the profile and Antonym, the controls for the processing environment; if the organization has not allowed any profiles and private mode is unavailable, there is no top row. The bottom row has, on the left, the conversation history fill ring, “Clear”, “Skills” and “Pinned notes”, and on the right, complexity and specialist.

“Skills” opens the “Conversation skills” window, and the pin button opens the “Pinned notes” window (see the sections above). The number in the corner of a button shows how many skills are connected and how many notes are in effect; at zero there is no number (the exception is a set changed by Iskra: the yellow mark is shown even at zero, including when the set has become empty), and more than nine is shown as “9+”; the exact number is in the button’s tooltip. The skills button has a yellow outline when you defined the set yourself, and gets a yellow mark when Iskra changed the set in the last reply, until you open the window. The notes button exists only in a personal conversation; if the notes could not be loaded, the button stays without a number and is labeled simply “Pinned notes”.

The ring is green up to 70%, yellow up to 90% and red above that; there is no number inside. On hover or focus, a two-line tooltip pops up: “Context is N% full” and “Click to compact now”. The ring itself is the compact button; there is no separate “Compact” button. The percentage shows how much of its allotted budget the conversation history takes up. The budget is calculated from the working context window, the amount of history the platform supports at an acceptable cost and response speed, so it is smaller than the model’s full window and does not change with the number of documents Iskra has read within the current task. When there is no measurement, the first line of the tooltip is “Will recalculate after the next

reply”. While the conversation has no messages, there is no ring at all, but “Clear” remains. While compaction is in progress, the arc spins, the tooltip says “Compacting…” and “The beginning of the conversation is being condensed into a short recap”, and “Clear” is unavailable with the tooltip “Compacting…”: only one operation at a time runs on the conversation context.

“Clear” is an eraser icon without a caption; its name appears as a tooltip, and the action asks for confirmation. This is the only place with context operations: they have been removed from the “Chat Statistics” window, which now has only the indicator. What exactly compacting and clearing do is described below in “Compacting and clearing the context”. Both operations keep pinned notes: only your action in the “Pinned notes” window removes them. In scheduled conversations, and on the start screen while there is no conversation yet, there is neither the ring nor “Clear”.

The parameter controls are square buttons without a border; the tooltip of each one names its current value. Clicking a control expands the parameter card above the footer; you can close it with the cross, the Esc key or by clicking outside it. The whole group appears only once the panel has loaded the conversation settings: while loading is in progress or if it has failed, there are no parameters in the footer. From the keyboard: press Tab to reach a control, Enter expands the list and places the cursor on the checked row, ↑/↓ move between rows, and Enter selects.

Profile is a chip with the name of the active profile in the top row; in the list, the organization’s profiles are shown as rows, with the selected one checked. If the organization allows exactly one profile, the list has a single row marked “set by the organization” and a footnote saying that there are no other profiles and that the administrator changes the profile. While private mode is on, the chip shows “Set by private mode”: the run does not use an organization profile, so there is nothing to choose.

Complexity is a button with the level name and a chevron: “Auto”, “Fast”, “Smart” or “Very smart”. If the conversation uses a level with increased consumption, the corner of the button shows a “×N” mark: token usage is counted at N times the base rate; at normal consumption there is no mark. In the list, each row is labeled and marked with a scale icon. “Auto” means that Iskra determines the level itself, choosing between “Fast” and “Smart”; “Auto” never chooses the maximum level, and under the list there is the line “Very smart — manual only, ×N charge”, which also states the price in words. If, because of images in the conversation, a level would lead to a model that cannot recognize them, that level is dimmed, with an explanation under the list. On platforms where automatic detection is turned off, the first row is “Chosen by the system” instead of “Auto”: it is checked until you choose a level yourself, and it hands the decision back to the platform.

Specialist is a robot icon, with the letter “A” in the corner when Iskra itself chooses the specialist. The list has an “Auto” row and all specialists; under the list there is a single description: for the row under the cursor or keyboard focus or, if there is none, for the selected one. On a phone, where there is no hover, the “i” button on a row shows the description. What a specialist is and how it differs from a skill is covered in “Specialist: a response role, not a skill” above.

Antonym is a shield in the top row: whole when private mode is on and crossed out when it is off. The expanded card has two segments, “Disabled” and “Enabled”: this is private mode, in which the request is processed on Iskra’s servers and is not passed to an external model provider. The request text does not change. While you have not chosen anything, both segments are muted and show the mode from your settings, and the line under them says “As in settings”; clicking a segment pins the mode for this conversation, and clicking the checked segment again removes the pin and hands the mode back to your settings. You can turn the mode off only if it is not set by a security policy; turning it on is available where the administrator has opened the mode to your organization (see below). If the policy is on, switching is unavailable, and the reason is shown under the switch: “Enabled by the security policy” (or “Set by the organization” when the mode is set for the entire platform). In that case, you can turn the mode off only in the security settings, either your own or

the organization’s. What the mode covers and what stays outside it is described in the Security and data privacy section.

The panel does not show the model itself: you cannot choose it, since it follows from the profile and complexity. If Iskra continued on a fallback model after a failure, the panel does not report that either; the charge multiplier on the complexity button shows the price of the model that actually produced the reply. A reason appears above the footer if a setting could not be applied, for example, if the selected profile was revoked.

In the same place, Iskra explains why a request went to a different level than the one you chose:

“The fast model is busy right now — the request runs on the regular one”: you pinned the “Fast” level, but the fast server is busy right now. This is not a failure: a stronger model answers, and usage is counted at the multiplier of the model actually called.

“The request runs on the regular model: the chat contains images the fast model cannot see”: the conversation contains pictures that the fast model does not see, so the regular model handles the request.

There is no separate “Back to automatic” button in the footer: to hand the decision back to Iskra, use the “Auto” row in the complexity or specialist list, and for the profile, the organization row.

The availability of private mode depends on your platform: if the administrator has not opened it to your organization, turning it on results in an error; turn the mode off or contact your administrator. If Iskra’s servers are busy, no reply arrives, and a message with a “Retry” button appears in the feed: the request was not processed, and you are not charged for it.

Working folder and connected Memory

The chat working folder (the chat’s system-managed folder or a selected Memory folder; see “Chat working folder” above) receives message attachments and files created by the agent; in the request panel, its contents are shown in the “Chat files” area. Besides the working folder, you can connect any other Memory folder to the chat as context: this does not make it the working folder, the chat does not work in it, and the agent’s files still go to the working folder. Connect it via “@ Folders” by typing “@” right in the input field, or with the “+” button in the header of the “Connected folders” area; a folder already chosen as the working folder for this chat cannot additionally be connected as context. You can connect several folders and set the agent’s access mode separately for each: “Read only” or “Allow changes” (for details, see the “Memory” section). Skills are connected separately, in the “Conversation skills” window, and are not mixed with connected folders.

Web search

Iskra can search the internet for up-to-date information and include what it finds in its answer.

If Memory folders or project files are connected to the chat, the assistant first looks for the answer in them, browsing the connected folders and reading the relevant document, and only then turns to the internet for whatever the folders do not have. This way, company regulations from a connected folder are not replaced by a generic answer from the web. To start with the internet, say so directly: “search the web first, don’t open the folder yet”.

To turn web search on or off, use the globe button in the message input area: a crossed-out globe means that search is off. The name of the state appears as a tooltip on hover. When web search is on, the assistant can query search engines for fresh information.

Sources. If an answer contains data from the internet, a “Sources” block appears under the assistant’s message. Each source includes:

The source site’s icon.

The page title.

A link to the original.

Site content and design. Usually Iskra takes the main text from a page and discards the markup and styles. Images from the main content stay marked in their places along with their available descriptions. Iskra looks at the ones it needs for the answer; looking at them does not create a file in the working folder. A file appears only if the image needs to be saved, downloaded or inserted into a document.

If you are interested in the look, such as brand colors, fonts, logo or layout, say so directly: “look at how the site is designed”, “make a presentation in the style of this site”. Iskra will then also read the page source code along with its stylesheets.

How replies are displayed

In the browser chat, the assistant’s replies support:

Headings, lists, quotes and tables: standard Markdown formatting.

Emphasis and links: bold and italic text, strikethrough, regular links.

Line breaks: a single newline is displayed as a line break, without a blank line between paragraphs.

Code blocks: displayed with syntax highlighting for various programming languages. Each code block has a copy button.

Math formulas: LaTeX via KaTeX, as display formulas $$...$$ or \[...\] and inline formulas \(...\). A single $...$ is not used (it is the currency sign).

Images: if the assistant has attached an image, it is displayed in the reply. Click the image to open an enlarged view.

Websites, phone numbers and email addresses in replies.

Website addresses in Iskra’s replies are clickable, even without https:// and www.: for example, example.com or a domain name written in a non-Latin script. The site opens in a new tab, and the chat stays where it is. Before you go to an external site, Iskra shows its address and asks you to confirm opening it. The same warning applies to links in the list of sources under a reply.

The “Do not show this warning again” checkbox is remembered after you click “Open website”, for your account in this browser. Cancelling or closing the window opens nothing and does not save the checkbox. When you view a public chat without signing in, the choice is saved in this browser.

Clicking a recognized phone number launches the system calling app, and clicking an email address starts a new message in your mail app. The external site warning is not shown for these actions. The available apps and further steps depend on your device settings.

Links to Iskra files keep the built-in preview or download. Addresses and phone numbers inside code blocks stay plain text.

If a technical problem occurs during generation, Iskra can continue the same reply on another model. The text written before the switch is not lost, and the reply continues. There is no separate notification about the switch, and model names are not shown to the user.

Special Obsidian extensions do not get dedicated formatting in the chat:

a > [!NOTE] callout is displayed as a regular quote with the [!NOTE] marker visible;

==highlight==, [[Note]] and ![[file]] stay plain text instead of becoming a highlight, a link or an embed;

the footnote syntax [^1] / [^1]: text is not supported, and how it is displayed is not guaranteed.

For predictable results, use regular quotes, bold text, links and attached images.

Reasoning blocks. Some models show their reasoning in collapsible “think” blocks. They are collapsed by default. Click the block header to expand it and see the model’s internal reasoning.

Actions on selected text and quotes

Select text within a single message, and an action toolbar appears next to the selection. On a phone, you do the same with a long press and the system selection handles. A plain selection does not create anything by itself: you can copy the text with Ctrl + C (Cmd + C on a Mac) or with the “Copy” button in the toolbar, and no form opens.

On Iskra’s replies, the toolbar offers “Quote”, “Copy” and, for a short single-line fragment, “Pin”. “Quote” puts the selected fragment into the input field: quotes are stacked and numbered above the message text; you can remove any of them with the cross, and clicking a quote shows its place in the reply.

While Iskra is replying, the toolbar works on finished text blocks that are already followed by the next block. The last part, which is still being written, is not available yet: if the selection includes it, the toolbar does not open. A pause to work with a tool does not by itself make the last block finished. After the reply ends, you can select any fragment of it again; quotes and pinned notes you have already added are kept. A toolbar opened during generation may close when the reply finishes. In an archived conversation and in read-only mode, these actions are unavailable.

A quote keeps the text exactly as you selected it. If the formatting changed as the reply continued (for example, a link appeared in place of its Markdown notation), clicking such a quote scrolls to the reply, but the exact fragment may not be highlighted.

“Pin” saves the selected line of up to 1,000 characters as a pinned note of the conversation, using the same mechanism as the /pin text command. You can view and delete notes in the “Pinned notes” window.

On your own messages, the toolbar offers “Copy”, “Quote in reply” (insert the fragment into your text as a quote) and “Pin” (make a short single-line fragment a pinned note of the conversation, like the /pin text command); the list of pinned notes and unpinning are in the “Pinned notes” window. The toolbar does not appear for a selection that spans two messages.

Sending quotes

A quote is a message in its own right. If the input field contains at least one quote, the send button is active even when the text is empty: Iskra receives exactly the selected fragment, with no made-up text, and replies to it. Your comment on the fragment is the message text itself: type it in the field, and it is sent together with the quote. An ordinary empty field with no quotes, files or voice recording still cannot be sent.

Quotes are sent with your next message, including through the queue while a reply is being generated. If the server did not accept the message, the quotes stay in the input field. Once the message is accepted, they appear above your message in the feed, and they stay there after a page reload too. If the reply stream then breaks off, the “Retry” action sends the same snapshot of quotes; once the retry is accepted, these quotes disappear from the previous draft. Quotes are not shown in a public link to the chat or in an export: only the author of the chat sees them.

If the original reply has changed in the meantime, Iskra does not send the request and offers to remove the outdated quote and select the fragment again; the message text and the other quotes stay in the field. You can add up to 8 quotes to one request; each fragment can be up to 4,000 characters, and up to 16,000 characters in total.

Message actions

When you hover over an assistant message, an action bar appears:

Copy — copy the message text to the clipboard.

Like / Dislike — rate the reply. This helps the system improve the quality of replies. Clicking again removes the rating. On some Iskra installations, the administrator has turned ratings off; in that case these buttons are not shown in the bar.

Hide — hide the message from the chat history. Hidden messages are not displayed but are kept in the system.

Managing chats

Each chat in Iskra is a separate conversation with the assistant. The system automatically gives the chat a title based on the first message.

Older chats load automatically as you scroll down a fully expanded list in the sidebar. If some chats are hidden by a collapsed section or a short project list, automatic loading is replaced by the “Show more” button: in the “Projects” section it expands chats that are already loaded, and at the bottom of the sidebar it loads the next part of the history.

Renaming. Hover over a chat in the sidebar, click the “⋯” (three dots) button that appears and choose “Rename conversation”, then enter a new name. On mobile, the “⋯” button is visible on the open chat without hovering. The top of the row menu has “Star”, “Rename conversation”, “Move to organization” and “Move to archive”; below the divider are “Chat Statistics”, “Share chat” and “Download chat archive”. You can also rename an open chat from the “⋯” menu next to its title at the top of the screen: it has the same items in a different order — first “Chat Statistics”, “Share chat” and “Download chat archive”, then below the divider “Star”, “Rename conversation”, “Move to organization” and “Move to archive”.

Moving to an organization. If moving is enabled on your platform, open the “⋮” menu of a regular chat in your personal profile and choose “Move to organization”. Select the organization and confirm the action — the move cannot be undone. After the move, the chat disappears from your personal profile and becomes available in the profile of the selected organization. The command is available only for chats outside projects and only if you have an organization available to you. A chat that works in a personal Memory folder stays with that folder; to work in the organization, start a new chat there. If other personal Memory folders are connected to the chat, disconnect them first.

Deleting a chat. Deletion is not permanent: the chat is archived and can be restored. How to open the archive and bring a chat back is described below in “Chat archive”.

Chat archive: archiving and restoring

To delete a chat, use the context menu in the sidebar. The chat is not deleted permanently but archived: it is hidden from the main list but kept in the system. An archived chat records the deletion date, its input field is locked, and a “This conversation is archived” banner with a “Restore” button appears below the message feed. When you restore the chat, the deletion date is cleared and the chat returns to the active list. To open the archive, click “⋮” to the right of the “Chats” heading and choose “Archive”; you can also restore chats there from the context menu.

If a public link was created for the chat, archiving automatically closes access through it: the page at the old address no longer opens for anyone, including the people you sent it to. The confirmation window warns you about this in advance. Restoring the chat from the archive does not bring the link back — if needed, share the chat again and Iskra will create a new address. Deleting the chat from the archive also closes access through the link. While the chat is archived, its “Share” button is unavailable: restore the chat first.

Public link to a chat

You can share a chat via a unique link. On a wide screen, click the “⋯” (three dots) button next to the chat title at the top of the screen and choose “Share chat” (in the “⋯” menu of a row in the sidebar, this item is available only for the open chat); on mobile, open the “⋮” (three dots) menu at the right edge of the header and choose the same item in the bottom sheet. The “Share public link to chat” window opens. Anyone who follows the link can read the chat and open the attached files (without being able to edit them), and they do not need to sign in to Iskra. The link leads to a published snapshot of the chat; the chat’s internal address (/conversation/...) does not grant public access. Once the link is revoked or the chat is archived, the page stops opening. If preview is not available for a format, the original file can be downloaded.

When you create the link, the “Show my name and avatar” option is available: if it is on, your name and avatar are added to the published snapshot of the chat, and readers see who shared it. If the option is off, the link stays anonymous — your name and avatar are not saved. By default, the value set in your profile settings is used. The toggle stays available after publishing too, so whenever you update the content, you always see exactly what readers will be shown.

You can share a chat even while Iskra is still generating a reply: the link includes only the part of the chat completed by that moment, and the reply in progress is not included — the “Share public link to chat” window warns you about this. To share the new reply as well, click “Update contents” once generation is finished: the link address stays the same, and its content is updated.

How to close access. If you no longer need the link, open the same “Share public link to chat” window and choose “Close public access”. After you confirm, the page at this address stops opening, and the attached files become unavailable. This action cannot be undone: the old address is not restored. If you share this chat again later, Iskra creates a new link with a different address, so there is no point in sending out the old one.

Chats in an organization profile. The organization decides whether its chats can be opened via an external link. By default, the “requires approval” rule applies: you create the link yourself, but nothing opens through it until the organization makes a decision — a reader sees that no such page exists. The organization’s Owner or Administrator can review the link and approve it; once approved, the address works as usual. If the organization has prohibited external links, Iskra does not create the link and tells you so.

The organization approves the content it has reviewed. After “Update contents”, the updated snapshot waits for review again, while the address stays the same. If the organization later prohibits external links or changes the rule, earlier approvals stop applying: the issued addresses stop serving content until a new decision is made.

Older organization links. Links to chats in organization profiles that were created before this rule was introduced are suspended: the old addresses do not open until the organization reviews them. Such a link is not revoked — once approved, it works at the same address, and you can close access yourself at any time.

If the chat belongs to an organization profile, Iskra also closes the link automatically when you are removed from the organization, leave it yourself, or the organization is deleted entirely. The old address no longer lets anyone open the chat, continue it or download its attachments. Closing the link does not delete the original chats, messages and files; however, once you leave or are removed, the organization profile becomes unavailable to you.

All your public links are collected in one place: Settings → Public links. There you can see which chats are published, when each link was created and when its content was last updated; from there you can also open the chat or the public page, copy the address, and close access to a single link or to all of them at once. This section is always available, even if creating new links is temporarily disabled on the platform.

If you move a chat to an organization, its public link follows it: the address does not change, and the link can be managed in the “Public links” section of the organization profile. However, the link stops opening at the old address — the chat has become organization material and is subject to the organization’s external access rule. To make the link work again, click “Update contents”: the snapshot is sent for the organization’s review, and once it is approved, the old address opens again.

Downloading a chat. The download button in the top bar of the chat builds a ZIP archive with the whole chat:

chat.html — the chat for reading in a browser: messages, the “Files” and “Starred” tabs, and the full timeline of every reply. The collapsed “Timeline” block keeps the assistant’s reasoning, tool and skill calls with their arguments, a short summary and the full output, subagent work, web search and Memory sources, the skill scenario checklist and status notices (reply interrupted, failure). The “Expand entire timeline” button expands all blocks at once.

chat.json — the same data in machine-readable form: useful if you need to analyze the chat or move it to another tool.

files/ — the entire contents of the chat’s working folder. For a chat with a system-managed folder, these are its own files; for a chat working in a selected Memory folder, it is that whole folder, including files created in it by other chats. Connected (non-working) Memory folders are not copied into the archive.

Files from the “Drafts” section are not included in this archive.

Drafts

If you started writing a message but did not send it, the text and the quotes you collected are saved automatically as a draft of that chat. The next time you open the chat, you see the text in the input field and the stack of quotes above it.

When you navigate within the current tab, Iskra also keeps the selected local files in the draft of the corresponding chat or project. After a full page reload, you need to select those files again; the text of a regular chat is restored from the saved draft.

In the sidebar, chats with an unsent draft are marked with a special indicator, so you can easily find unfinished conversations.

Sidebar

The sidebar on the left contains the general “Chats” list. When there are matching chats, separate “Starred” and “Skill chats” sections appear above it. Each section can be collapsed separately. In the “Chats” heading menu, you can choose the order:

Single list — regular chats grouped by time.

Projects — project chats are moved to a separate section; the six chats already loaded are shown first, and the “Show more” button opens the rest.

The single list uses these time groups:

Today — chats from the current day.

This week — chats from the last 7 days.

This month — chats from the last 30 days.

Older — all other chats.

On a wide screen, you can collapse the sidebar to a narrow strip of icons with the button in the top right corner of the sidebar. To expand it again, use the same button at the bottom of the narrow strip, below your avatar. The “Chats” button in this strip opens a compact list of chats; you can close it with the Esc key or by clicking outside the list.

Creating a new chat. Click the “New Topic” button at the top of the sidebar or use the keyboard shortcut Ctrl + Shift + O (Cmd + Shift + O on Mac).

Each chat in the list shows its title. Chats with unread messages (for example, messages that arrived while you were in another chat) are visually highlighted.

Searching messages and chats

Click the magnifying glass button next to the “Chats” heading to search the messages of the chats available to you. Type at least two characters, and the results update automatically. Search tolerates one typo; exact matches are shown before fuzzy ones. If you open search from a chat, matches from the current chat take priority within their group. Clicking a result opens the chat and highlights the message found.

Chat statistics and context

On a wide screen, click the “⋯” (three dots) button next to the chat title at the top of the screen and choose “Chat Statistics”; the same item is in the “⋯” menu of a chat row in the left sidebar, so you can also open statistics for a chat that is not currently open. On mobile, the item is in the “⋮” (three dots) menu sheet at the right edge of the header. When the monthly token limit is used up, a red dot appears on the three dots in the header, and the item itself is highlighted in red.

The same window opens with the /usage (or /tokens) command sent from the input field of an existing chat, without calling the model. You can also ask Iskra itself about the usage of the current chat (“how many tokens have been spent in this chat?”): it reads the same statistics and reports the input tokens, the output tokens and the total; if the data is temporarily unavailable, it says so.

The “Chat Statistics” window shows:

Conversation history — how much of its budget the chat history (what the assistant “remembers”) takes up. The budget is a share of the working context window, not the model’s entire window: tool results within the current task do not count toward it. The indicator is green below 70%, yellow from 70% to 90% and red above 90%. The value is taken at the time of the last request and updated after the assistant replies. It counts the saved summary and the remaining active history: the savings on an individual request, described below in “Context compaction and clearing”, do not change the percentage — otherwise the number would “jump around” just because part of what was found did not go into one particular request. The same indicator and controls are in the footer of the request panel.

Chat tokens — how many tokens went to your requests (prompt) and to the model’s replies in this chat, their sum in the “Total tokens” row, and the counted usage in megatokens (MT). If a higher power level was turned on for part of the work, the MT total already includes the multiplier shown by the “×N” mark on the “Complexity” button. This is the accumulated usage of the current chat at the moment you open the window: completed model calls are counted, and the most recent ones may appear in the statistics with a delay of a few seconds.

Monthly limit (or the organization’s shared pool) — how many tokens have been used in the current subscription month across all chats and the agent, and how many remain until the limit resets. The month is counted from the payment date, not from the first day of the calendar month, so there may be several days left until the reset.

Context compaction and clearing

The longer the chat, the more of the history budget its history takes up. For this, the footer of the request panel has a fill ring (clicking it compacts the context) and a “Clear” button. They appear in a personal chat once it has been created, including a personal project chat. On the start screen, before a chat exists, there is nothing to manage; scheduled sessions have no context operations at all.

Compact — the assistant briefly summarizes the early part of the conversation and continues the chat with this summary. The last full turn is not summarized, so requests like “shorten this answer” still refer to the original text. The meaning of the conversation is preserved, and space in the context is freed up. A dotted “Context above was compacted” divider appears at the boundary: messages above it are given to the assistant as the summary, and messages below it stay verbatim. If there is nothing to compact yet and only one short turn remains, Iskra tells you so without calling the model again.

Clear — the chat starts from a clean slate: the assistant “forgets” all the history above, and the messages themselves are removed from the feed and no longer show up in search. Only the “Context cleared” divider at the start of the chat remains of the previous conversation. The action cannot be undone, so the button asks for confirmation in a separate window. If you need the history, compact the context instead of clearing it, or save the history in advance: the removed messages are not included in the chat export either.

You can run the same actions right from the input field: send the /compact (compact) or /clear (clear) command as a regular message; /clear also asks for confirmation. While a reply is being generated, context operations are unavailable. Both operations keep pinned notes: they describe the rules of the conversation, not its history (see “The ‘Pinned notes’ window”).

The assistant can also clear the context itself if a skill instructs it to (for example, clearing the context after each processed document). In this case the feed is kept: only what the model sees leaves the context, while the conversation stays on screen with a divider at the boundary.

Compaction runs automatically after a reply or before the next request if the history no longer fits in its budget. Iskra first saves a summary of the early turns, then continues working. The recent part of the chat stays verbatim as long as it fits in the context; a very large completed turn can also go into the summary. The “Context above was compacted” divider appears in the feed. Until the next reply, the ring shows “Will recalculate after the next reply”. While compaction is in progress, sending a new message may be temporarily blocked. Compaction keeps a brief summary; the original conversation and files remain available.

If a required compaction fails, the reply does not start. The history and files are kept. Repeat the request or click the context fill ring to start compaction manually.

Iskra stops resending large results from older turns to the model in full: in the working request, each is replaced by a note with the tool name (and the size, when it is known exactly), and if needed the assistant rereads the data with the same tool. Recent results and anything that fits into a short output are sent verbatim. These results are still fully taken into account in the summary and during compaction, so what was found earlier stays in the chat.

Within one long task, Iskra tidies up on its own: it removes previously read results of file and search operations from the active context and, if needed, briefly condenses the beginning of the work and continues from that note. What is removed is not lost — the assistant returns to the fragment it needs when it needs it again, and the reply still relies on the facts found.

When you start typing with “/”, Iskra shows the slash menu: in an existing chat, the top section is “Commands” with /compact, /clear, /pin (“pinned notes”: without text, it shows and unpins notes; with text, it pins one) and /usage (in scheduled sessions, only /usage remains among the commands, since context operations are unavailable there); below it is the “Scenarios” section (ready-made scenarios of your skills, if you have any), and under that the skills themselves that are available for explicit invocation (one line per skill). Keep typing part of a name, and Iskra filters the matching scenarios and skills; you can also search by the name of a tool inside a skill. Move through the list with the ↑/↓ arrow keys; confirm the highlighted item with Enter or Tab (or a mouse click), and Esc closes the menu. A skill that is missing some setup also stays in the list, marked “not configured”: invoke it as usual, and Iskra asks for the missing data itself.

Notifications

The bell icon in the top bar opens the list of notifications. Each row shows a title, the time and an icon on the left that indicates how important the notification is. A gray dot on the right means the notification has not been read yet; the counter on the bell itself counts exactly these. The list also keeps read notifications — they stay in place, just without the dot, so anything you viewed recently does not disappear. Click a notification to go to the related action or chat; the “Mark all read” button clears all the dots at once without deleting anything.

Pop-up notifications in the corner of the screen are the same entries. The cross removes such a notification from the screen but does NOT mark it as read: the dot in the list stays, and the notification does not pop up again. The list holds fifty entries, and unread ones take precedence: when space runs out, the oldest READ entry gives way, even if it is newer than some unread one. So the list is not an archive: notifications viewed long ago are pushed out over time. The counter shows the exact number of all unread notifications, including those that did not fit in the list: for example, 63. After you read five notifications, it drops to 58. “Mark all read” also marks entries outside the list as read.

Overview

Agent mode extends what a regular chat can do by giving the assistant a set of tools for practical tasks: writing and running code, working with files, searching the web and other operations.

What an agent is

An agent is an AI assistant that not only answers questions but also takes actions. Unlike a regular chat, an agent can:

Run code — write and run programs in various programming languages.

Work with files — create, read and edit files.

Search for information — query search engines and external resources.

Use tools — call specialized tools to solve tasks.

The agent works in an isolated environment, which keeps code execution secure.

How agent mode works

In the current chat, Iskra handles practical tasks in agent mode automatically. There may be no separate mode switch: if a task requires tools, the assistant starts the necessary agent turn, and the interface shows tool calls, questions and result files right in the chat.

The availability of individual tools depends on your plan, the platform settings, connected skills and your permissions in the current profile. The agent can always create files in the working folder of the current chat — the chat’s system-managed folder or a selected Memory folder (see “Chat working folder” in the “Chat” section). It sees another Memory folder only after that folder is connected to the chat, and changes it only if write mode is allowed.

The agent sees connected folders and refers to them by their regular names, for example “Contracts 2026”, just as in the interface, not by technical identifiers; the same name is shown in the tool call header (for example, read path=/Contracts 2026/act.pdf). If text has been extracted for a document (see “Memory” → “Extracted text”), the agent’s search and read tools may show an auxiliary file with the .text.md suffix next to the original. This is a technical representation of the extracted text for the agent: it is not shown in the “Memory” section and is not a separate file; in its reply, the agent links to the original document, not to this file.

As long as at least one folder is connected to the chat, on every turn the agent first looks through the connected folders (ls, grep, read) and only then goes to the web (web_search, fetch) for what is not in the folders; an answer to a question that a document from a folder covers is based on that document. A direct request to start with the web changes this order.

Reading files

Iskra opens a file from the chat’s working folder or a connected Memory folder, or a file attached to a message, with the read tool. A plain text file is read as is, regardless of its extension: .txt, .md, .csv, .json, as well as files with uncommon extensions such as .sql, .svg or .sh. Iskra can read a text file in a different encoding if the encoding is declared in the file itself; otherwise it says so and offers to re-encode the file. A non-text file is available if the platform can parse it: Word, Excel and PowerPoint documents, PDF, EPUB, FB2, DjVu and ZIP archives. Iskra cannot read a binary file of an unknown format and says so.

Iskra reads a PDF with a text layer directly. If the pages are scans, Iskra recognizes them with the image model configured by the administrator. So whether a separate specialized recognition tool is available does not determine whether Iskra can read a scan.

An office document and an HTML page each have two representations, and Iskra chooses between them based on the task. By default, it takes the content: the document text, the values and formulas of spreadsheet cells, the page text without markup. This is enough when you need information from the file. When the formatting matters, Iskra reads the same file in its source form: the HTML styles and layout, or the structure and formatting of the office document. Iskra makes this choice itself based on the task, so you do not need to specify the representation. A page that renders itself with a script, such as a dashboard or an interactive report, does not

contain any text of its own; Iskra opens such a file in its source form and keeps working with it as usual.

That is why a request like “make a report styled like this file” with an attached HTML sample can be fulfilled: Iskra takes the real styles from the sample, not just its text. If the formatting matters, say so directly, and Iskra picks the right representation right away.

Reading large materials

You do not need to split a large document into parts manually in advance. Iskra does not pull the whole file into the context at once, but how it reads depends on the file type.

Iskra reads a plain text file (and HTML in source form) by ranges: the read tool takes the number of the first line and the number of lines, and Iskra finds the range it needs by searching the file. If the answer needs a summary of the entire content, Iskra first compiles that summary with a script into a smaller file and then reads that file.

A document that goes through extraction (PDF, Word, Excel, PowerPoint, EPUB, DjVu, ZIP) works differently: the platform returns the extracted content in full, and specifying a line range has no effect on it. Even so, a large result usually does not land in the reply all at once: the platform holds it and gives it to Iskra in parts, which Iskra reads one after another. The exception is the rare case when the extracted text itself exceeds the platform’s technical limit: then reading is not split into parts but ends with an error stating the exact limit, and the document needs to be reduced or split.

When there is a lot of material — several documents, long pages, a series of search queries — Iskra delegates reading and sifting to subagents and asks them to return the facts it needs with references (see “Subagents within a single task”).

Separate processors and their limits. The size Iskra can read is set not only by the general upload limit. Word, Excel and PowerPoint office documents are processed by a separate converter with its own, noticeably smaller size limit. A new office attachment larger than 50 MiB is rejected in the chat before its content is transferred. A previously saved file may also turn out to be too large to read; Iskra reports the allowed size. The document needs to be compressed or split, and for reading it can be exported to PDF. Reading page by page does not help here, because the limit applies at the processor’s input. Other supported formats have a higher input limit that matches the platform’s general file limit, but there is also a second limit, on the amount of extracted text: a file can pass the size check and still turn out to be too large after extraction. The recommendation in both cases is the same: reduce or split the document. These are limits of specific processors, not a refusal to work with large documents.

Audio and video transcription

Where the administrator has enabled the platform transcription tool, Iskra transcribes supported audio and video files from the chat’s working folder, a connected Memory folder or project files, including common video formats, not just audio. If the tool is turned off in the installation, Iskra says so directly instead of promising a transcript.

A recording is processed in full: one file, one transcript. While processing is under way, the card shows the current stage, the elapsed time and the recognition activity. This is information about the work, not a completion percentage. A long wait does not by itself mean an error; Iskra keeps waiting for the same task. Once the transcript is ready, the full text received from the speech recognition model automatically appears as a TXT file in the “Drafts” section of the current chat. Iskra reads this file and prepares the result you need: a brief summary, meeting minutes or a formatted transcript in the working folder. The raw recognition text remains available in the draft so you can check against it.

You can upload a file of up to 500 MiB with an audio track of up to 3 hours. A closed installation may have a stricter limit on the size of processed audio. Exceeding a limit produces a clear error; the platform does not cut the recording into pieces or trim it to the allowed size. If the model could not finish the transcription, Iskra reports this without automatically repeating the paid recognition. A large text is stored in full in a file; Iskra can read the parts it needs and search it. If the file could not be saved to the draft, the completed task can be retrieved again without a new recognition.

In the transcript, utterances are split by voice: one continuous stretch of speech can span several sentences. Timestamps mark the start of an utterance approximately; for an exact quote, check the original recording. The same voice may get different labels in different parts of a long recording, so the number of labels does not by itself indicate the number of participants. If saved labels are available, Iskra receives them along with the text, including when it returns to an already completed transcript.

Saving the model’s entire output does not guarantee that it recognized every word: check important fragments against the audio. If recognition detected no speech, Iskra reports this; if speech is audible, this may be a recognition error. The separate built-in “Audio transcription” skill works only with audio and small files; it is a different mechanism (see the “Skills” section).

Creating office documents

For a new document, presentation, spreadsheet, dashboard, image, audio or video, Iskra takes into account the entire required result and the format you named. When editing an existing file, it also takes into account the need to keep the file itself and its format. First, Iskra checks the connected skills available for this request, including those not yet used in the chat. It chooses the most specialized of the skills that meet all these requirements. The selected skill determines the work steps, the format and the number of result files. The search for a suitable skill also applies to tasks that do not create a file, for example requesting information from 1C, taking into account the skill’s configured mode of operation.

If the built-in “Presentations” skill is selected for a presentation, Iskra picks a design for the topic and audience itself and prepares an interactive HTML presentation. It creates a print version with PDF when you ask for print or PDF, and both versions when you ask for them together. An uploaded PDF used as a content source does not change the format of the result. Details are in the “Built-in skills” section. If there is no suitable skill and no format is specified, Iskra creates a .pptx with the built-in Office tool. If you need a PowerPoint file specifically, name the format explicitly: “PowerPoint”, PPT or PPTX. A suitable skill that supports this format gets priority; if there is none, Iskra creates the file with the built-in Office tool. Word documents and Excel spreadsheets are handled the same way. During the work, you may see the tools create_presentation,

Iskra converts Word, Excel and PowerPoint to modern formats with the same export.

The “Presentations” and “Dashboard” skills pick a design for the data and the task themselves, without choosing a theme first. To continue work started by a skill, Iskra uses that skill again if it suits the requested edit, and takes its working draft into account.

For a factual result, Iskra checks the relevant connected sources and available public sources. If you requested a structured file or template and the named source is unavailable or contains only part of the data, Iskra first creates a draft from the information available. It keeps the subject you named and marks missing fields, for example as “no data” or “needs clarification”. After the draft, Iskra explains what information is needed to complete it and asks you to attach, provide or connect a source.

In a regular fact check, a missing detail stays unconfirmed; Iskra names the data that is missing for an answer. Realistic fictional values are used when you choose the suggested option with demo data or explicitly ask for a demo or prototype. In that case they are clearly marked both in the material itself and in Iskra’s reply.

If there is no suitable skill and no format is specified, a new plain text document (a report, policy, meeting minutes, instructions) is saved as a single Markdown file, and a plain table as a single CSV file. You can get another format through the download menu. If you named a format explicitly, Iskra follows that requirement, including for plain text in .txt.

When you need DOCX specifically and the built-in Office tool is selected, Iskra usually writes the content in Markdown and converts it to .docx. Both files remain in the chat’s working folder: the Markdown file as the source for edits and the finished Word document. Iskra switches to step-by-step assembly via create_word_document where Markdown is not enough: merged cells, complex table geometry, page breaks at specific places, image placement, a special title page.

In this Office path, the document gets built-in formatting: three levels of headings, captions, quotes, tables with a highlighted header row, margins and page numbers. If the installation administrator has uploaded a custom Word template, that template is applied.

When exporting an office document to PDF, Iskra first finalizes the editable .docx, .xlsx or .pptx source and then exports it via export. Both files remain in the chat’s working folder: the source for further edits and the finished PDF. For the result of a selected skill, the set of files is determined by that skill’s workflow.

Iskra converts a text source (Markdown, HTML, plain text or CSV) to PDF directly with the same export, without an intermediate Word file. For such a PDF, you can set the print layout right away: orientation (portrait or landscape), paper size (A3, A4, A5, Letter, Legal) and margins in millimeters, for example “make this PDF landscape” or “A5 with 10 mm margins”. Parameters you do not set are taken from the built-in PDF template or the one uploaded in the organization, and the formatting of text and tables is preserved. For a PDF made from Word, Excel and PowerPoint documents, the page geometry is taken from the document itself: to change it, first adjust the page settings in the source file via

Iskra converts an attached DJVU or DJV file to PDF with the same export. The conversion is done by the platform server, the same one that prepares the PDF version for preview and the download menu, and the finished .pdf appears in the chat’s working folder next to the source, so you can keep working with it.

Operations on finished PDFs

You can split an attached PDF into separate pages or ranges, merge several PDFs into one, and compress a file (for example, a heavy scan).

Reverse conversion from PDF to DOCX is not supported. PDF is a target export format, not a source for an office document: office formats are converted only within their own family, while Markdown, HTML and plain text are converted to .docx, .odt and .rtf and back. If you need an editable document from a PDF, Iskra tells you plainly that there is no such conversion and offers instead to extract the text and build a new document; the original layout will not be preserved. Iskra performs these operations with local tools in an isolated sandbox without network access: the document is not sent to external online services and does not leave the platform. The result appears in the chat artifacts as separate files. Compression lowers the resolution of images inside the PDF, so the gain is noticeable for scans, while for “text-only” PDFs the size may barely change; in that case Iskra tells you so plainly.

Iskra can insert into an office document an image that the agent already has access to: for example, a logo attached to the current chat or a file from an accessible connected folder. An arbitrary path to a file on your device or in a folder you have not granted access to cannot be opened: attach the file or grant access to it first.

Iskra inserts official symbols (coats of arms, emblems, seals, logos) from an attached or accessible local file that the user has identified as the current official mark. An exact edit of such a file (for example, cropping it from a letterhead, removing the background, scaling, rotating or changing the margins) is marked as derived, with the source file and the operation indicated. The resulting version is considered official only if the user or an official source has separately confirmed that specific version. If only a link or the text of a page is available, Iskra immediately creates the requested draft with a labeled placeholder for the mark and asks you to attach the exact official file. A version drawn from a description is marked as an approximation.

If the administrator has not configured a custom PPTX theme, you can ask for one of the built-in themes: neutral light (blank), blue (blue), brown (brown), dark (dark) or gray (grey); without an explicit choice, blank is used. All built-in themes are structured the same way; only the palette differs. Each has six ready-made slide layouts: title, title and content, section header, two content, title only and blank. That is why you can change the theme even after the presentation has been created, and the design is rebuilt without redoing the slides. The administrator’s custom theme is applied automatically instead of a built-in one. A new presentation gets its design, but the theme’s demo slides and content are not copied into the user’s file.

Iskra does not overwrite an existing file when creating a presentation, Word document or spreadsheet. To change a file that has already been created, ask Iskra to edit that specific file. If the requested name is taken by a file in the Trash, the new file may get a free name with a numeric suffix.

Processing spreadsheets with Python

For bulk analysis of large spreadsheets, Iskra uses the same local Python in an isolated sandbox: it reads and aggregates data from xlsx, ODS and parquet files and calculates statistics and summaries over volumes that are impractical to go through manually. This path is for analysis, not for a finished document: Iskra still creates and edits a formatted spreadsheet for the user with its standard tools (see “Creating XLSX spreadsheets” and “Editing office files” below). The analysis result appears in the chat artifacts as a separate file, for example a CSV.

Files in connected Memory folders are visible to such a script at the same paths and under the same names as in Memory itself; for example, Python can directly read a spreadsheet from the connected “Contracts 2026” folder. Access is read-only: the agent changes Memory with its own tools, not from a script. Iskra saves the result of the script separately, in the chat artifacts.

Creating XLSX spreadsheets

Ask Iskra to create a spreadsheet in XLSX format. In your request, you can explicitly specify the number of sheets, their names, order and purpose. Iskra keeps this structure when creating the file or tells you plainly if a requirement is not supported yet.

Skill materials

You can link Memory folders with materials to a skill: a corporate presentation template, a document letterhead, last year’s reference report, a coat of arms or a seal. Put the files in a Memory folder and link it in the skill card. You only need to do this once, not in every chat.

When the skill is invoked, its folders become visible to the agent as read-only. The agent works with a copy of the material: it takes the file, puts it in the chat’s working folder and edits that copy, while the originals stay untouched.

This covers three typical tasks: making a document from a corporate template, taking last year’s report as a reference and reworking it for new data, and inserting a coat of arms or a signature into a document.

For a personal skill, the owner links the folders. For an organization skill, the owner or an administrator does it once for all employees; the folder then needs organization-level access, otherwise colleagues cannot read it. A skill from the Catalog comes with the shared resources of its bundle. Links to the author’s Memory folders are not transferred: users connect their own materials in their current profile. Examples of choosing where to put files and checking the result are in the skills guide.

The agent sees the entire contents of a linked folder, so it is better to keep skill materials in a separate folder rather than mixed in with other files. If the folder is deleted or access to it is revoked, the card marks the material as unavailable. The owner or an administrator can unlink such a row and choose another folder.

Editing office files

Writer documents (including .doc, .docx, .dot, .dotx, .docm, .rtf and OpenDocument), Calc spreadsheets, Impress presentations and Draw drawings are edited via edit_office_document. The tool works with files in the chat’s working folder (the chat’s system-managed folder or a selected Memory folder) and with connected folders that allow changes.

Iskra keeps the original path and format of the file. Editing runs in an isolated environment with no access to the network or to the other files in the working folder.

If the source file was changed in parallel, Iskra does not overwrite the new version and reports a conflict. If an error or timeout occurs, the original version also stays unchanged.

After a significant creation or edit of a text (Writer) file, Iskra checks the final layout page by page via review_page: the tool returns a PNG of each page for visual review, Iskra fixes noticeable clipping, overlaps and other defects, and before finishing it reviews all pages of the current version once more. These images are used only for internal checking and do not appear among the result files. A structured post-edit snapshot for Writer is not returned to the agent.

The page-by-page check shows the document the way LibreOffice renders it. In Microsoft Word, complex layouts, non-standard fonts and certain OOXML elements may look slightly different.

Legacy .doc, .dot and .rtf documents can be edited directly via edit_office_document. To get a modern format, Iskra uses the same export and creates a sibling file next to the original (.docx/.dotx) without changing the source; you can then keep working with that file via edit_office_document. Conversion works only into a new path. If the sibling file already exists, this is a final conflict: use that file, and do not delete or rename it to convert again. If it is in the Trash, first restore it from the chat files section. A legacy file with macros, an embedded object or other active content may be rejected during conversion: the platform does not remove such content silently.

The legacy .ipf format cannot be opened or edited directly. If you ask Iskra to work with such a file, the only way it can get the content is the explicit convert_legacy_presentation conversion. For an IPF file of up to 10 MiB, it creates a permanent sibling .pptx without changing the source file and without adding

the result to the old message. The conversion never overwrites an existing PPTX: Iskra uses the sibling file that has already been created. If it is in the Trash, first restore it from the chat files section, then repeat the request to read or edit it.

Tool calls

When the agent decides to use a tool, the chat shows information about the call. Each call appears in real time: you see which tool the agent is using right now, without having to wait for the reply to finish.

While Iskra is still preparing the call parameters, the timeline shows a separate preparation card without service arguments. When a Word document is being created or edited, the card

shows a plain-text fragment of the document script being generated. The same view is used for other Office create/edit calls with a program. When the call is ready, the card is automatically replaced with the regular tool block. If preparation broke off because of an error or an interrupted stream, the unfinished step stays in the timeline with a corresponding mark.

If no new results arrive for more than 10 seconds during long processing, the indicator shows the messages “Still working — the task is in progress” and “This is taking a little longer”. This means the task is continuing and you do not need to send your message again.

Each call is shown as a compact two-line block:

Header — the tool name and key parameters (for example, read

a spinning indicator while the call runs, a green check mark on success or a red cross on error.

Result preview — a short summary under the header (for example, “1 line read” or “3 results found”).

If a call has detailed output, you can click the header: this expands the full result text in a monospaced block (scrollable when the output is large). Clicking again collapses the block.

Tool calls are shown in chronological order within the agent’s reply. This lets you follow the course of the work and see what steps the agent took.

This same list is also a way to check the agent’s report. If the reply says that a file was read, copied or saved, the corresponding call must be in the timeline. The platform checks this itself: a reply that describes work with files but contains no calls at all does not complete the task. Iskra gets a reminder that no actions were taken and either actually performs them or corrects the reply in the same message.

Skill cancellation and timeout

Skills (tools from the skill catalog) can take noticeably longer than regular tools. For such calls, the card shows additional information:

Countdown — if the skill has a time limit, the remaining time is shown next to the status while it runs (“~Ns remaining”); as the limit approaches, the countdown is highlighted. Cancel button — a running skill has a “Cancel” (✕) button on the right. Clicking it actually stops the skill (down to its isolated sandbox) rather than just hiding the card. After cancellation, the agent receives the result “stopped by user” and continues working.

Timeout completion — if a skill exceeds its time limit, the card ends with a timeout mark instead of spinning forever. A cancelled skill is marked accordingly.

Working folder limit

The agent’s working folder holds up to 1,000 files per sandbox operation. If an operation exceeds the limit, its result is not saved, and the agent pauses the task and asks a question that states the limit (for example, “the operation exceeded the file limit — limit: 1000”): “Stop” means the agent finishes the reply based on what has already been done; “Continue compactly” means it continues by editing existing files. Your files are protected: once the limit has been hit, the agent cannot delete existing files until the end of the task. If extra files really do need to be deleted, tell the agent which ones: deletion is possible only on your explicit instruction. Separately from the file count limit, there is a time budget for saving results: if one operation creates or changes many hundreds of files in a single run, some of them may not be saved in time. The agent sees this in the operation report and can redo the work more compactly.

Continuing a long task

A task started in the chat keeps going without asking whether to continue and without being forcibly ended just because of total token usage, total time or the number of steps. This also applies to subagents. You can stop the task in the chat; clarifying questions, required confirmations, loop protection and limits on individual operations stay in place.

The monthly usage limit is checked when a new request is sent. If it runs out during a task that has already been accepted, Iskra can continue the task, and all further usage is counted. API calls, scheduled tasks and apps keep their own limits: the API has 90 seconds for a synchronous turn and 15 minutes for a background run; apps have a prepaid run limit; scheduled tasks have a limit on concurrent runs and are stopped if a run hangs without progress. These 90 seconds and 15 minutes do not apply to scheduled tasks.

If you close the working folder limit question with the cross or do not answer it, the agent does not break off mid-sentence: it wraps up the task and gives a complete reply on what has been done, with the results of the subtasks already finished and an explicit list of what remains. You can continue with a regular message such as “Continue”.

Hard interruption remains a fallback: if the agent still fails to finish the reply, it keeps the part already done, and a message about the interrupted execution appears below it. The intermediate result and the created files stay in the chat. If the interruption happened before the agent managed to write the final reply, an error card with the reason is added to the message, for example that the model did not respond in time. Such a turn is not presented as completed, and after a page reload it looks the same as at the moment of the failure.

If an unfinished intermediate turn without a final result was saved, only “Continue” is available: it sends a new request in the same chat with the saved turn in the context, rather than repeating the interrupted request. If such a turn could not be saved before any actions were performed, the request can be retried. A standalone skill run collapses into a result line once it finishes and does not offer to continue a deleted intermediate turn; if needed, run the skill again. A reply that consists only of a created or delivered file is considered successfully completed and offers neither of these actions. If a file has already been created but the reply could not be saved or linked to it, Iskra separately warns you not to repeat the run: the file may have

been saved in the working folder, and a repeat may run the same generation again and charge for it again.

Visual view from a skill

While running, some skills show data visually right in the call card: as a table, a list, “key-value” pairs, status badges or text blocks. This makes the progress easier to follow than raw text.

This view is live: it is shown while the skill runs and is not kept after the reply finishes (the skill’s result stays in the card’s regular output). The skill builds the view itself from a fixed set of safe elements; arbitrary HTML or scripts cannot be displayed in it.

Input form from a skill

Some skills may ask you for data while running, for example to choose a city or a plan, or to confirm an action. In this case the skill pauses, and a form appears right in the call card: one or more fields (text, multiline text, a single choice from a list, multiple choice, a “yes/no” checkbox) and a submit button.

Fill in the fields and click the button named after the action, such as “Send”, “Install” or “Delete skill”; the skill receives your answer and continues with what you entered. A confirmation form can also have no fields: then it contains only a description of what will happen and a button. The time to respond is limited: if the form stays unfilled for too long, the skill ends with a timeout. The form is temporary: it is shown only while the skill is running.

Running a skill scenario

When a ready-made skill scenario (a sequence of predefined steps) is running, a live progress checklist appears above the input field: the scenario name, a “done / total” counter and a list of steps with the status of each one (in progress, done, error, skipped, pending). The “Hide” / “Show” button in the header collapses and expands the list of steps; it only changes the view and does not affect the scenario itself.

The assistant carries the scenario through to the end in the same reply: it closes all the steps (done or skipped) and sums up. If execution was interrupted earlier, for example because of a reply timeout or a dropped connection, two buttons appear under the checklist:

▶ Continue — sends a command to continue the scenario from where it was interrupted;

Finish — forcibly ends the scenario: the assistant confirms the cancellation and briefly sums up what was and was not done.

For an unfinished scenario, this is the only “Continue” button: the message about the interrupted reply stays in the feed as an explanation but does not duplicate the action under it.

After completion, whether normal or forced, the checklist above the input field disappears, and the reply itself keeps a final line that stays in the chat history (and in its export) permanently: “Scenario ‘Name’ completed” with a list of steps, or “Scenario … cancelled” (with the reason, if there is one). A completed or cancelled scenario can be run again only explicitly, with the button on the skill card, the scenario selection chip or a direct link; it does not resume on its own.

If the previous reply was interrupted by a timeout or a dropped connection and the scenario was left unfinished, Iskra continues it in the background on its own with your next message on the scenario’s topic, so you do not need to click “Continue” separately. If you ended the scenario

yourself with the “Finish” button, there is no automatic continuation: this is a deliberate decision, not a failure, and you can start over only by running the scenario explicitly.

How Iskra handles complex engineering tasks

Iskra takes into account the available text, attachments and the conversation context. In automatic mode, the level can rise from “Fast” to “Smart” and then stays there until the end of the conversation. A short follow-up after a complex task does not lower it. When data is missing, Iskra chooses “Smart”. You can choose a different level manually; switching back to “Auto” restores the automatic level already reached. Model availability, context size and images can affect which model produces the response. Usage is determined by the step actually performed and the multiplier shown.

Specialists are selected automatically at the start of a conversation and kept for subsequent messages; you can change them manually. A specialist role does not raise the power level by itself: in a pair, the primary specialist defines the way of working, the secondary one adds expertise, and the response is still a single one. Iskra keeps refining the set of skills to fit the task, regardless of the selected specialists.

For non-trivial engineering tasks — for example, multi-step data processing or code made of several interdependent parts — Iskra runs the super-iskra process: first it drafts a plan, sends it for independent review, then implements it and checks the result again. It assigns the middle of the process — drafting the plan, reviewing it, implementing it and reviewing the implementation — to separate subagents; running the result to check it, debugging and analysis are always done by Iskra itself, so the verification is not independent. If a subagent run ends with a technical error, Iskra performs the middle of the process itself as well and warns that there will be no independent reviews — the order of phases and the file artifacts are still preserved. The process does not start for simple questions and short scripts. The super-iskra process is not the only way to work with subagents: it merely adds a fixed “plan — review — implementation” order on top of them, and Iskra can also delegate work to subagents in an ordinary agent turn (see “Subagents within a single task”).

While subagents are working, a separate strip above the message box shows their statuses: how many are working right now, what each one is doing and for how long — if a subagent stays silent suspiciously long, the strip highlights this separately. The process materials — the plan, reviews, run reports — are saved as files in the plans/ folder inside the “Chat working folder” in the request panel, so the whole course of work stays visible and verifiable.

Iskra does not switch to a higher power level on its own, though — you choose the model level. If a task looks heavy, Iskra will remind you in a separate line where this choice is: the “Complexity” button in the bottom row of the request panel; the strongest level is “Very smart”. The selected level applies starting with the next message, and the button shows a “×N” mark — the token usage multiplier of the model that will produce the response; the explanation is in the expanded complexity row. If the response had to be continued on a fallback model, further steps are counted at that model’s multiplier. Meanwhile, the subagent strip above the message box shows the progress of the task.

On borderline tasks, where it is not obvious whether the full process is needed, Iskra will ask you whether to apply it. If you do not need the process at all, you can turn it off on the “Skills” page by switching the super-iskra skill mode to “Off”.

How to clarify a vague request: the “Prompt” skill

When a request is hard to put into words, the “Prompt” system skill can help. Its slash command is /prompt (always in Latin letters: this is the skill’s technical name, while “Prompt” is the title of its card in the Catalog). Type /prompt in the message box or choose the skill in the “/ Skills” menu — Iskra checks what is missing for reliable execution across four dimensions, “Context / Goal / Task / Expected result”, asks one compact

round of questions only about what is missing, shows a short working brief and immediately performs the task. The skill does not re-ask what is already clear from the request and the chat history; if you do not answer the questions, Iskra continues with explicitly stated reasonable assumptions or names the specific reason why it cannot continue.

Occasionally Iskra brings in the skill on its own — only when a request genuinely cannot be started reliably; a clear, actionable request is executed right away, without questions. “Prompt” is a lightweight clarification of the wording; it does not replace the super-iskra process for complex engineering tasks. Like other skills, it can be turned off on the “Skills” page.

Result files

The agent publishes the finished files it creates as a result of its work — for example, a document, a spreadsheet, a chart or the output of running code — in the “Chat files” area of the request panel. Intermediate sources, scripts and not-yet-verified versions may first appear in the collapsed “Drafts” section below the working files. After verification, the agent moves exactly the finished version to the working folder; the draft remains available separately for viewing, history and further targeted edits within its retention period.

When preparing images, diagrams and slide sets, Iskra checks the result against the assignment: the number of files, required elements, text readability, cropping and overlaps. Contradictory requirements, such as a different number of slides in the request and in the plan, need clarification first. The check applies to the final version: after edits, the image has to be reviewed again. If viewing is unavailable or covers only part of the result, Iskra must say what was left unchecked. Such a check helps catch errors but does not guarantee flawless layout or accurate generation.

Actions with result files:

View — click a file row to open its contents in a modal window. Text files are shown with syntax highlighting, and spreadsheets in an Excel-like read-only mode with sheets, merged cells, formulas, drop-down lists and static PNG charts. Spreadsheets are not converted to PDF. Other actions are collected in the “⋯” menu next to the file row (on devices with a mouse the button appears on hover; on touch devices it is always visible): download (with a format submenu if a PDF version is available), copy the contents of a text file to the clipboard, version history — a list of previous versions from which you can restore any of them (see “File version history” in the “Chat” section) — and delete.

The area header holds the “Starred” and “Trash” icons (they appear when there is something to show) and, when the folder contains files, the Memory icon “Go to Memory”, which opens the chat folder in the “Memory” section; the header has no option to download all files as an archive.

Previous file versions are useful to Iskra itself as well: if a file is corrupted or overwritten by mistake, it looks at the file’s history and restores a suitable version instead of re-creating the file from scratch. Restoring works the same way for Iskra and for you — the previous content becomes a new version, and the history is preserved.

Separately, the platform protects against accidental loss of content: if a script processing an existing file suddenly shrinks it at once both by more than a tenth and by more than a few kilobytes, the file is not overwritten — Iskra sees a refusal with the exact “before / after” numbers and either confirms the reduction deliberately or fixes the command. The other files of the same step are saved as usual.

For XLSX spreadsheets, Iskra reopens the saved file after creating or modifying it and recalculates the formulas. If at least one cell still contains a formula syntax error, including an error from the source file, the new version is not published:

Iskra receives the cell address and fixes the formula. Calculation errors, such as division by zero caused by data that is still empty, do not block the file, but the result gets the unverified mark with a list of the affected cells.

If the chat is opened in the context of a project, the agent also uses the project’s files as context — read-only. They are not listed in the request panel: the project card in the first row of the “Connected folders” area opens the “Connected folders” window on the “Project” tab, which shows the project’s folders and “Project files”.

Placing a result on a board

You can show a finished file to people outside the chat by placing it on a board. If you ask the agent to do this (“post the report on the ‘Product launch’ board”), it only prepares the placement: it picks the board, the files and the versions. The agent cannot publish the material itself — the placement happens only after a person has confirmed it.

The agent hands the prepared placement over to you the same way a skill hands over its form: to confirm

you see exactly what will be placed: the board and its audience, the tabs, the version mode of each one and the board’s tab counter. The “Place” button performs the action once — reopening the same page shows the result that has already been saved and does not create a second tab. The page outlives the agent’s response: you can close it, sign in again and come back to the same address.

The page does not stay valid forever, and conditions may change before you open it: the confirmation deadline and outdated preparation are general placement rules, described in the “Boards” section. For the agent, only one thing follows from them: if the page is no longer valid, ask the agent to prepare the placement again.

Declining simply means closing the form: until you click “Place”, readers get nothing. Neither the model’s response nor an automatic confirmation by a skill counts as a placement. The agent also does not change the board’s audience or its external sharing: these are separate actions taken by a person. Details about boards, audiences and versions are in the “Boards” section.

Questions from the agent

While working, the agent may ask you a clarifying question or request permission for an action. While the question is waiting for an answer, its form replaces the message box — the place where you usually write to the agent. After you answer or stop the request, the regular message box returns.

You don’t have to answer right away. A clarifying question, a skill form and a request to set up a connection stay by the message box even after Iskra has finished its response: they survive a page reload, switching between chats and a break of several days. When you come back, answer the question, fill in the form or save the settings — Iskra will continue the same task on its own; you don’t need to repeat the original request. In the chat feed, this appears as your new message containing the answer. A pending step is kept for 30 days; once that period expires, the card disappears and a note that the wait has expired remains under Iskra’s response. Archiving or deleting the chat cancels its pending steps.

The question block contains:

Title and wording — a brief description of the situation the agent needs to clarify.

Answer options — ready-made options with descriptions. Choose the one that fits; you can change your choice before submitting.

Your own answer — an option with a built-in field for your text. A ready-made answer is marked in the list and is not copied into this field. A question without ready-made options can be left empty: the agent will treat this as “no data”. A question with ready-made

options requires a choice or your own text. To submit the whole form, click “Submit answer”.

Answer deadline — an ordinary question has none: its header says “You can answer later”. A countdown timer (in “minutes:seconds” format) remains only on the question about the working folder limit — it relates to the task running right now, and when the timer runs out, such a question is closed.

Actions at the bottom of the form — “Close” lets the agent continue without an answer, and the stop button ends the entire current request. When you stop, the question disappears and unfinished tools in the chat feed are marked as cancelled.

While the question can be answered, the tab title starts with the ⏳ icon. When a question appears, a short sound plays if the browser’s permissions and sound autoplay rules allow it. The icon disappears after you answer, close the card or the time runs out.

Permission requests (for example, to run code or work with files) look similar but have only two buttons — “Allow” and “Decline” — and are sent automatically when you click, without a free-text field.

After you submit your answer, the agent continues working with your choice in mind. You cannot choose again — the answer is recorded immediately. The completed block disappears from the message box, but a record of the question and its result remains in the chat feed: the history is not lost.

Skill forms

A skill may ask you to fill in a form — for example, email details or a document confirmation. While Iskra is working, the form opens over the chat. If you put it off, a “The … skill is waiting for a form” card with a “Fill in the form” button stays by the message box: the regular fields you have filled in are saved as a draft and restored after a reload. After you submit the form, Iskra calls the skill again, now with your data, and carries the task through to the result, for example by creating a file. The cross on the card cancels the wait: a cancelled form does not come back.

If a skill lacks connection settings, a “The … skill is waiting for connection settings” card with a “Set up” button appears by the message box — it opens the secure form of the relevant skill (for example, Telegram, not the setup wizard). Passwords and tokens are entered only in this form and never get into the chat. After you save, Iskra checks the connection and performs the original request on its own.

Subagents within a single task

In agent mode, Iskra can create subagents on its own within a single task — no separate permission or special mode is needed for this. It delegates independent parts of the work to them: research sections, reading large documents, analyzing and checking an intermediate result — and then combines the reports it receives into one response. Subagents work in parallel with each other, whereas regular tool calls in the main turn run one after another.

A subagent can view images and document pages, recognize text in scans, prepare visual files and check finished HTML using screenshots. Long HTML pages are checked with consecutive scrolling screenshots at desktop and mobile sizes. This is a visual check; whether buttons and filters work is checked separately. If the capture is incomplete, the review marks the limit of its coverage. A subagent works with the files of the current task within that task’s access rights; the main agent receives its report and assembles the final response. If a subagent could not complete its part, it returns a reason instead of a result: Iskra finishes that part itself, reassigns it, or states plainly in the response what was not done and why; a partial result is never passed off as complete.

This is a standard capability of agent mode. It should not be confused with parallel chats (see “Parallel tasks in different chats” below) or with the super-iskra process (see “How Iskra handles complex engineering tasks” above): super-iskra is a separate skill that adds its own “plan — review — implementation” order on top of subagents, but Iskra creates subagents without it as well.

Delegation most often kicks in where there is a lot of material: several documents, long pages, a series of search queries. In such cases, Iskra asks a subagent to read and sift through the material and return the needed facts with links, instead of pulling the whole text into its own context.

Subagents have limits. A subagent does not create its own subagents — the tree stays one level deep — and does not ask you clarifying questions: only the main turn asks them. Only a limited number of subagents work at the same time, so some assignments wait in a queue for a free slot. The total number of subagent assignments within one task is not limited; subagents have no separate MT usage limit. Their work counts toward the overall usage of the task and stops when the task is cancelled. Each subagent works for no more than 40 minutes after leaving the queue and makes no more than 150 tool calls. When it reaches either limit, it finishes its work and hands Iskra a partial report: what is already done, which files are saved and what remains to be done. Preparing the report may take about another minute. The main agent continues the task with the results it has received. Cancelling the task stops both the subagents and the preparation of their reports.

Subagents start working on the model configured for them, with reasoning. If a suitable model is unavailable, the main Iskra does the work with its own tools. Limits on individual operations and protection against repeating identical calls also stay in effect.

Parallel tasks in different chats

Agent tasks can run in parallel in different chats. If the platform is overloaded or agent mode is temporarily unavailable, a request may end with an error message; try again later or contact your administrator.

While one task is running, you can keep using the platform in other chats.

Overview

The Inspector is a protective check of messages in Iskra. Before your request is sent to the assistant, the platform checks it for obvious security risks — for example, an accidentally pasted secret or dangerous input. This check has no separate chat or portal: it works unobtrusively in the background.

What happens in your working chat

Regular assistant responses arrive in the current chat as always. If the Inspector finds a critical security risk in your message, what happens next depends on the strictness mode set on the platform:

Strict mode (default). The message is not sent to the assistant. It is marked as unsent, the warning text is shown below it, and the assistant does not receive this message in its context. You can send your next request as usual.

Soft mode (warning). The message is marked with a warning but still goes to the assistant — the check only draws your attention to the risk without blocking the conversation.

Which mode applies is determined by the platform administrator or your organization’s administrator.

The prompt-injection check (attempts to rewrite the assistant’s rules) works in soft mode by default: such a message is not blocked — it goes to the assistant with an additional protective instruction to answer the safe part of the request and decline the request to break or reveal the rules. A question like “which model are you” is not blocked either and gets a substantive answer, but it is recorded in the platform log as a detection, and the assistant does not name the specific model or provider. Ordinary conversation management — “forget my earlier instructions about style”, “ignore the previous message, here is the correct link to the conversation”, “reset the context” — is not considered an attempt to rewrite the rules: such messages are not stopped even in strict mode and get a substantive answer, and the assistant receives the same additional protective instruction along with them. The instruction applies only to the message that triggered the check and does not carry over to subsequent messages in the conversation. The check does not recognize every wording: a safe quote that looks similar may be stopped in strict mode, and a non-obvious attempt to rewrite the rules may reach the assistant with the protective instruction. The installation administrator can enable a strict mode in which obvious attempts to rewrite the assistant’s rules, such as “forget all your instructions”, are stopped before they reach the assistant.

What can trigger a warning

A warning appears if a message looks like a critical leak or dangerous input, for example:

real API keys, tokens and passwords;

personal data;

internal URLs or infrastructure addresses;

classified or internal-use markings.

Remove the sensitive data or replace it with safe examples, then send the message again.

Recommendations and learning

The initial learning profile is collected by the general onboarding wizard. The separate AI tutor, its chat and the learning profile page have been decommissioned; previously created Tutor chats may remain in your history as regular saved conversations. The Inspector does not depend on this and always checks new messages; the strictness mode determines whether it blocks the risk or shows a warning.

Overview

Projects (internally, the entity is called a space) are shared workspaces where several members use a common project context: instructions, files, the member list and personal chats within the project.

What projects are

A project brings a group of users together around a common topic, project or task. In the current interface, work inside a project happens in personal chats: the assistant takes the project’s instructions and files into account. The shared project feed has been decommissioned and is not available for new messages.

Each project has:

A name — a short description of the topic.

A description — a detailed explanation of the project’s purpose.

An icon — a visual marker for quick recognition.

Instructions — instructions for the assistant that define the context and its behavior in this project.

Context files — documents uploaded for the assistant to use.

Project list

To go to projects, use the corresponding navigation item. A page opens with a list of all projects available to you.

Projects are shown as a grid of cards. Each card contains the project’s icon, name and description. Click a card to open the project.

At the bottom of the page there is a button that switches to archive mode, which shows previously archived projects.

Creating a project

To create a new project, click “Create project” on the list page.

Fill in the form:

Name (required) — the name of your project.

Description — a short explanation of the topic or purpose.

Icon — choose a suitable icon from the set. More than 60 options are available, grouped by theme: documents, tools, science, nature, animals and others. A folder icon is selected by default.

Instructions — instructions for the assistant. Here you can specify how the assistant should behave in the context of this project: which language to respond in, what tone to use, which topics to treat as priorities.

Click “Create project”, and you will be taken to the new project.

Working in a project

Inside a project, the following are available:

My chats — personal chats in the context of this project. The assistant takes the project’s instructions and files into account, but by default the conversation is visible only to you (see “Shared access to chats” below).

Project files — the project’s canonical folder in Memory. It is automatically connected to the project’s chats within the granted permissions.

Memory — additional folders connected to the project. The project’s chats inherit them; a specific chat can only narrow the access mode.

Members — the list of people who have access to the project.

Instructions — instructions for the assistant in this project.

The assistant in a project. Create or open a chat in the “My chats” section and write a message. The assistant will respond taking into account the project’s instructions and uploaded files.

You can start the first personal chat right on the project page. Before sending, the same “@ Folders” and “/ Skills” buttons are available as in a regular chat. Your selection carries over to the chat being created and applies to the first message; if the transition fails, it stays in the draft.

Iskra in a personal project chat. No separate @Iskra mention is needed: send a message as in a regular conversation. Iskra takes into account the history of this personal chat,

the project instructions and the project files available to it. The response, the reasoning and tool results are shown in the same personal feed.

You can manage the context of a personal project chat with the same “Compact” and “Clear” buttons as in other personal conversations. If an important rule should apply in all of the project’s conversations, save it in the project instructions or in a project file.

Right panel. The right panel contains the “Files”, “Memory”, “Instructions” and “Members” sections. Section headers stay visible even when collapsed.

Project Memory folders. Only the project owner can connect folders and collections to a project, including the first connection with the “+” button. The “+” button in the “Memory” section of the right panel opens the same picker as “@ Folders” in the chat: the folders available to you and a separate “Collections” section listing your own collections (collections cannot be opened to other users, so other people’s collections do not appear in the picker). Members without the right to connect do not see the “+” button — they see the same list of connected items, but without the mode switch and without the disconnect button. A connection can be narrowed to “Read only” or disconnected — the switch is right on the connection row; disconnecting does not delete the source files or change the collection’s contents. Connected folders and collections become context for the project’s personal chats. If a folder is deleted or access to it is revoked, it stays in the list as “Unavailable folder” and is not used as context, and the project owner can disconnect this link. The same happens to a connected collection if its owner has deleted it — the row stays as “Unavailable collection”. See the “Memory” section for details.

Members. The list of project members is shown in the “Members” section of the right panel. To add a new member, start typing their name or email in the search field — the system will suggest matching users.

Project settings. The project owner can edit the name, description, icon and instructions right on the project page. To change the name or description, click the corresponding field.

A project can also be selected for an entry in “Plans”. See the “Plans” section for details.

Shared access to chats within a project

Each personal chat created within a project is visible only to its author by default. If the author wants to show the chat to other project members, they can turn on shared access — a toggle in the form of an “eye” icon appears next to the chat in the “My chats” list:

Private chat (crossed-out eye icon) — the chat is visible only to the author.

Open chat (open eye icon) — the chat is visible to all project members, but only the author can write in it; for everyone else, the message feed opens in read-only mode, with the “This chat is read-only” banner shown below it.

Only the author of the chat can switch the mode. If the project has been moved to the archive, switching is blocked.

Project vs. board: how they differ

A project is a shared workplace: the chats, context files and instructions its members use. A board is a separate page that displays selected files; files are placed on it one by one, and the project’s chats, context files and instructions are not visible to the board’s readers.

Shared access to a chat, described above, opens the conversation to project members inside Iskra. It is not a board: a board has its own set of tabs, its own owner and its own

audience.

Being a project member does not give access to a board: a board is opened by its own audience — the owner, the whole organization, a selected organization group, people invited by email address, or anyone who has the secret link. That is why a file from a project stays private until the board owner chooses an audience themselves. Audiences and placement rules are described in the “Boards” section.

Navigating between project chats

When you are inside a chat linked to a project, a navigation bar of the form “project icon / chat name” is shown at the top of the window. This is the project breadcrumb:

Clicking the project icon and name takes you back to the project’s main page.

Clicking the chat name opens a drop-down list with the project’s other chats. Each chat in the list shows its name and last update time; the current chat is highlighted. The same drop-down list also has a “New chat” button that creates a new chat inside the current project.

This lets you quickly switch between chats of one project without going back to the project list page.

Context files

You can upload files that will become part of the project context. The assistant will take the contents of these files into account when answering questions.

To upload a file, click the upload button on the project page. You can also paste text from the clipboard as a file: use the “Paste text” button and specify the file name and contents.

File search is in the right panel, directly below the upload and paste-text actions. It filters the current file list without opening a separate dialog.

Uploaded files are shown as a compact list. Normally you see the file names; size, download and delete appear when you hover over a row. Click a file to open a preview.

Previewing project files works the same way as in chat: PDF, DJVU, images, Markdown, code, spreadsheets, Word/Office documents and presentations open right in the window. Iskra shows DJVU, supported documents and presentations as a server-generated PDF version. In the download menu, you can choose Original or, for non-spreadsheet formats, PDF if available.

If the browser cannot open an image, for example HEIC, Iskra tries to prepare a compatible copy for viewing. Downloading the original saves the source file.

XLSX, XLS, XLSM, Excel templates, ODS/FODS/OTS, CSV and TSV open in a read-only spreadsheet mode, without conversion to PDF. It supports sheets, copying, merged cells, formulas, formatting, frozen panes, drop-down lists and static PNG charts. Editing, interactive charts, pivot tables and sparklines are not supported. Beyond 200,000 non-empty cells, the preview becomes partial and Iskra shows a persistent note; the original remains available for download.

Next to the “Files” heading, file storage usage is shown in the “used / limit” format along with a small circular indicator. Clicking it opens a window for upgrading to a higher plan if you need more space.

Project files are subject to your plan’s file limits: the maximum size of a single file and the total file storage of the personal profile. This is the same

storage that is used for files in chat; there is no separate limit on the number of files in a project. Current values are shown in the subscription settings and on the pricing page.

Before an upload starts, Iskra checks whether the file fits within the size limit and whether there is enough total file storage. If not, a window appears with an explanation, a comparison of your current plan with the recommended one and, when an upgrade can solve the problem, a button to proceed to payment. If you need to free up space without changing your plan, delete unneeded files from the project or from chats.

The platform administrator can also set a technical upper limit on file size for the entire installation. In that case, the actual single-file limit will not exceed the technical limit, even if your plan allows more.

Archiving and restoring

If you no longer need a project but want to keep its contents, archive it.

Archived projects are hidden from the main list. To view the archive, click “Archive” on the project list page. You can restore any project from the archive, returning it to the active list.

Overview

In Iskra, the user model has two levels: account and profile. This lets one person work in several contexts — personal and in one or more organizations — without mixing data.

Account and profiles

Your account is your Iskra user account: email, password, name, avatar, interface language and theme. A person has only one account, and you sign in with it. Personal security settings (changing your email or password, deleting the account) live at this level.

A profile is a working context. Every account has:

A personal profile (always exactly one). It is created at sign-up. It holds your personal conversations, projects, documents, token usage and subscription for private use.

Organization profiles (from zero to several). They appear when you are invited to an organization. Each such profile is isolated: its conversations, projects and token usage are kept separate from your personal ones and from other organizations.

All your data — conversations, projects, files, statistics and subscription — is tied not to the account as a whole but to the active profile. That is why switching context changes what you see in chat, in the sidebar and in settings.

Profile switcher

On a wide screen, the active profile is shown in the top right corner: an avatar (yours for the personal profile; for an organization, a circle with its initials), the name and a chevron; clicking it opens the list of profiles and the “Create Organization” item. The same profile is shown at the bottom of the sidebar: for the personal profile, your avatar, name and plan; for an organization, a circle with initials, the organization name and your role. From there, the switcher opens via the profile menu: click your name, then the “Profile” item. On mobile, the switcher (a circle and the workspace name; a long name

is shortened to 9 characters) is in the header on the right, next to notifications, and opens the list from the bottom; the profile menu has the “Profile” item there too.

On your first sign-in, Iskra automatically opens the profile of an available paid organization, if you have one. If there are several paid organizations, the one you joined first is chosen. After sign-in, your explicit choice in the switcher applies for the current session and does not change because of a page refresh or onboarding.

Each browser tab keeps its selected context independently: you can work in an organization in one tab and in your personal profile in another at the same time. Switching the profile changes the context only in the tab where you did it; other open tabs keep working in their own profiles, including after a page refresh. A new blank tab opens in the most recently selected profile. If you duplicate an open tab, the copy first inherits its profile and then works independently.

The list of available profiles includes your personal profile and all organizations you belong to. Each organization shows your role (Owner, Administrator, Member). Blocked organizations are shown in the list, but you cannot switch to them.

When you select another profile, the interface reloads in the new context: the list of conversations, projects and documents, the subscription indicator and the monthly limit all change. Personal data does not leak into the organization, and vice versa.

The same menu has a “Create Organization” item — any user can set up their own organization (access to its working features opens after you pay for an organization subscription; see Settings → Billing).

What an organization is

An organization brings several profiles together under a common name: a company, a department, a study group, a project team. An organization provides:

unified management of members and their roles;

internal groups for finer structuring (departments, study cohorts, business lines);

a shared subscription under which all members of the organization work.

You can be a member of several organizations at once — each of them appears as a separate profile in the switcher.

Organization page

You can open the active organization’s page via the navigation in the sidebar. The page consists of several tabs, some of which are visible only to administrators and owners.

Info. Basic details:

name and avatar (the first letter of the name if no icon has been uploaded);

description (if set);

identifier (the short name in the address);

statistics: how many members and how many groups.

Employees. A list of the organization’s members with their name, avatar, role and email. Above the list there is a search by name and email and a role filter (all, owner, administrator, member); when a filter is active, a counter of matching employees is shown next to it. The list with search and filter is available to regular members as well as to the owner and administrators — only the actions differ. The owner and administrators add and remove employees, change their roles, choose groups when adding someone, and change an existing employee’s group membership with the “Manage groups” action. In a member’s row,

they also turn on the “Allow sharing boards externally” right — more on it below, in the “External sharing of organization materials” section.

When an employee is removed, all public links to conversations of their profile in this organization are closed automatically. The same happens when a member leaves the organization on their own; when an organization is deleted, the links of all its profiles are closed. This closes public access but does not by itself delete conversations, messages or files.

Groups. Units within the organization. A regular member sees the list, the membership and the roles. The owner and administrators of the organization create groups, edit their name, identifier and description, and manage their membership. An administrator of a specific group manages its details and membership without having administrator rights for the whole organization.

Only the owner or an administrator of the organization can delete a group — the “Delete group” button is in the group editing window (it is not shown to an administrator of the group itself who has no organization administrator rights). Before deleting, Iskra asks you to confirm the action and warns about the consequences: members will be removed from the group, and its access to Memory folders will be revoked. Active invitation links to the group are not revoked — they can still be used to join the organization, but they no longer add new members to the deleted group. Only an organization administrator can re-create a group with the same name.

Subscription. The organization’s plan (package): the current plan, limits and payment. All members see this tab, but plan management — choosing a package, paying, buying additional tokens — is available only to the owner and administrators. See Settings → Billing for details.

Usage. A summary of the organization’s megatoken usage from the shared monthly pool, plus a row for each member: how much was used, the limit and its source — a “personal” quota or the “shared pool”. This is also where the owner and administrators manage personal quotas: the “Quota” button in a member’s row opens a window where you can set a monthly limit in MT or remove it (without a personal quota, the member draws on the shared pool). An administrator can also set a quota for themselves; the organization owner is not assigned a personal quota. The tab is visible to the owner and administrators when the organization package is active.

When a quota is used up. In the organization workspace, the chat is blocked, and the message states which quota has run out and does not suggest getting a personal subscription, since that has nothing to do with the organization. A member’s personal quota: “You have used up your token quota in the organization” — the owner or an administrator can increase it on the “Usage” tab. The package’s shared pool: the owner is offered to buy additional tokens, and for other members the message says to contact the owner. The organization plan’s monthly limit: the message is informational — the owner is told when the limit renews, and other members are told whom to contact.

Skills. This tab is available only to administrators and owners (and appears if the feature is enabled on the platform). Clicking it takes you to the “Skills” section. Organization skills are collected in the “Organizations” section of the “My skills” page and are managed there:

List of published organization skills — organization-level skills that already work for all its members. Clicking a card opens a page with details.

Skill update — you can submit a new version of a skill from the details window.

Skill revocation — you can revoke a skill from the same window. After confirmation, it is no longer available to the organization’s members; the list updates within a short time.

From the same tab, you can submit a new skill for publication to the organization. See the Skills section for more about working with skills.

Security. The tab is available only to the owner and administrators. Here you configure privacy for the whole organization — the “Antonym” private mode for its members: you can turn it on or off if the platform administrator has made the mode available to your organization. When the mode is on, members’ requests are processed on Iskra’s servers and are not sent to an external model provider. If the platform administrator has made the mode mandatory, the switch is locked and Antonym works for everyone. See Settings → Antonym for details.

Roles in an organization

Roles apply at the level of the profile in a given organization, not at the account level. The same person can be an Owner in one organization and a Member in another.

Role — Capabilities

Owner: full control — managing members, groups,

organization settings and the subscription. Can appoint

administrators.

Administrator: managing members on the “Employees” tab and

groups on the “Groups” tab; access to the organization’s skill,

usage and security settings.

Member: viewing information about the organization and the list

of employees and groups. Working in shared projects and

conversations.

A member’s role is shown as a colored label next to their name in the list.

External sharing of organization materials

Organization material goes outside by two different routes: a board is shared externally by a member who has been granted that right individually, while public links to conversations and an app’s public address are governed by the organization’s policy. The personal workspace is subject to neither, but organization material does not become personal just because it was placed from a personal chat.

Boards: a member’s personal right. A board’s external audiences — “Named recipients” and “Link” — are unlocked by the “Allow sharing boards externally” right. The organization owner and administrators have it by virtue of their role, while a member is granted it individually, with the toggle in their row on the “Employees” tab; by default, members do not have this right, and the board’s access window says whom to ask for it. The right is checked every time a board is opened, so revoking it immediately closes all of that person’s external boards to readers, without affecting their colleagues’ boards. A personal board does not require the right: people manage their own material themselves.

Chats and apps: the organization’s policy. The “External access” tab of the organization page, which only the owner and administrators can see, holds two settings: “Chats: inviting outsiders” and “Chats and Apps: public address”. For each one, you choose “Prohibited” or “Requires approval”. While “Requires approval” is in effect, a new public link to an organization conversation waits for a decision by the owner or an administrator and does not open until then; approval applies to a specific snapshot, and updating the content sends the link back for review. Links issued before the policy appeared go through the same review. An organization app in “Everyone” mode is subject to the same rule: while “Prohibited” is selected for the public address, the app does not open at that address.

Tightening the policy takes effect immediately: addresses already issued stop opening, including for a reader who has the page open, at the next check. Prohibiting either of the two settings invalidates the previously issued approvals for both of them: they have to be issued again, and allowing it again does not revive the earlier approvals. Boards are not affected: they do not read the external access policy, and “Prohibited” does not close them — whether a board is shared externally is decided by the member’s right.

How approving an app differs from sharing a board is described in the “Apps” section. What boards are, what audiences they can have and how a secret link is issued is described in the “Boards” section.

Organization apps

If apps are enabled on your platform, the “Apps” section works in the context of the active profile. An app of your personal profile does not appear in an organization profile, and the apps of one organization are not mixed with the apps of another. When you switch profiles, the list of apps, the available actions and the management context change. When you open an app, Iskra will separately offer you a choice of profile if several suitable options are available to your account.

Permissions depend both on ownership of the specific app and on the profile’s role in this organization:

The app owner publishes and activates its versions, chooses the access mode, and stops, starts and deletes the app. If the owner is a regular Member of the organization, they submit the request for public access themselves, and a specific version must be approved by an organization Owner or Administrator who does not own the app.

An organization Owner or Administrator sees all of the organization’s apps, including those in “Owner only” mode. For someone else’s app, they can set the “Owner only” or “Organization” mode, stop, start and delete the app, and approve an explicit request from its owner. They cannot initiate the move of someone else’s app to “Everyone” mode themselves. Nor can they activate a version of someone else’s app.

A Member sees other people’s organization apps in “Organization” and “Everyone” modes but does not manage them. They manage their own app as its owner.

“Everyone” mode means access for everyone, including unauthenticated users. An app owner who has the organization Owner or Administrator role confirms the warning once and turns on public access immediately. An app of a regular Member needs two separate consents: first from the app owner when making the request, then from an organization Owner or Administrator who does not own the app when approving it. Each window opens with the checkbox cleared; a previous consent does not carry over to the next attempt.

Approval applies to a specific version, and a new version of a public app requires a new approval. An app owner with the organization Owner or Administrator role activates the new version and approves it on their own in a single action. For an app of a regular Member, the saved request allows an organization Owner or Administrator who does not own the app to explicitly approve a ready version that is not yet current. Once the warning is confirmed again, this version becomes current. If the app is stopped, it must be started before a version can be activated or approved.

Emergency notifications about the run state changing to “Unhealthy” or “Failed” go to whoever owns the app at the moment of the change, not to all members of the organization. A repeat of the same state does not duplicate the notification; after recovery, a new emergency transition creates a new one.

Inviting members

The owner or an administrator of the organization adds a new member in one of the following ways.

By email. On the “Employees” tab, enter the future member’s email. The form will not accept an address with a typo — for example, an extra bracket or a space inside it — or with non-Latin letters before the “@”: no chip appears, and a hint below the field says what to fix; no invitation is created for such an address. If the person already has an Iskra account, the organization will appear in their profile switcher right after their next sign-in. If they don’t have an account yet, they will receive an invitation to sign up, after which they automatically join the organization.

Next to each address, the email’s status is shown: “pending”, “sent”, “sent (delivery not tracked)”, “delivered”, “send failed”, “delivery failed” or “accepted” — if the person has already used the invitation.

The “Send again” button creates a new email with the same link. If the previous email has not gone out yet, the recipient may see two invitations — this is normal, as the link in them is the same.

By invitation link. Create a link (optionally tied to a specific group) and send it any way you like.

When opening an invitation link, the user sees:

the organization’s name and avatar;

the group name (if the invitation is tied to a group);

the “Join” button.

To join, you need to sign in to Iskra. If the user is not signed in, the system offers to sign in or sign up and then brings them back to the invitation. If the user is already a member of the organization, the system says so.

After the user explicitly clicks “Join” and the invitation is accepted, Iskra opens this organization as the active working context. If the organization temporarily fails to open, click “Retry”: processing an already accepted invitation again does not create a duplicate membership.

How this relates to subscriptions

Subscriptions in Iskra are also tied to the profile, not to the account:

personal plans are paid for on the personal profile and apply only there;

an organization package is paid for by the owner and covers all members of that organization;

a member of several organizations uses the subscription of each of them in that organization’s own context.

An organization package provides a shared monthly MT pool, and the number of members is not limited by a number of seats. The recommended number of users for a plan helps you choose the package size but does not block invitations.

For more about plans and payment, see Settings → Billing.

Overview

If you have a promo code for an organization subscription, only the organization owner can activate it (the Organization → Billing section).

There are three types of benefit:

Bonus megatokens — add N MT to the organization’s shared pool for a set number of billing periods.

Free period — gives N days or N billing periods of subscription without payment.

Discount % — reduces the package price for the next N payments.

How to activate:

When purchasing a package — enter the code in the “Promo code” field on the checkout page. The price or terms are recalculated automatically.

On an active package — in the organization’s billing section, use the “Activate promocode” field. A discount applies to subsequent payments, bonus MT are added to the pool, and a free period extends the package.

Active promo code card (visible only to the owner): benefit type, remaining periods, validity period and a “Cancel” button. A free period is applied once and cannot be cancelled from the card. Bonus MT can be applied in addition, but the same code can still be redeemed by an organization only once.

Discount when paying by invoice. A percentage promo code applies when you buy or renew a package, whether you pay by card or by invoice. For a first purchase, enter the code before choosing “Pay by invoice”; an already activated code is applied automatically. A discount period is used up once the payment is confirmed; issuing, cancelling or expiry of an invoice does not use it up. If the organization has a permanent discount, the larger of the two discounts applies; they are not combined.

Important:

A discount and a free period are mutually exclusive: if one such promo code is already active, you cannot activate a second one.

Bonus MT are additive, but the same code cannot be activated again.

A given code can be redeemed only once — after cancellation, it cannot be activated again.

Upgrading to a higher package and buying additional tokens are charged at full price unless the promo code terms explicitly apply to the chosen action.

Legacy promo codes for free or bonus seats do not apply to organization package plans.

Possible errors:

“Organization promocode not found” — check the spelling (a personal promo code does not work on an organization subscription).

“Promocode is unavailable (expired or disabled)” / “Activation limit reached”.

“The organization already has an active promocode” — cancel the current one.

“This promocode was already used by the organization”.

“Promocode is not valid for the selected plan” — the code was issued for other plans.

“Benefit type does not apply to the package” — the promo code belongs to the old per-seat payment model.

Overview

The settings section lets you manage your profile and security, subscription, file storage and application preferences, and view usage statistics. In the expanded sidebar, click the gear icon; in the collapsed sidebar, click your avatar, then “Settings”.

Settings are divided into tabs: Profile, Security, Audit, Statistics, Storage, Public links, Application. In addition, if the corresponding features are enabled on the platform, the Billing, Referral Program, API Keys and Learning Profile tabs appear (the last one only when the learning feature is enabled).

Much of the data is tied not to the account as a whole but to the active profile (see the Account, profiles and organizations section). That is why the subscription, statistics, token usage and some profile settings change when you switch context with the profile switcher in the sidebar.

User

Here you can view and edit your account details. The section has the same name in the settings menu and in the profile menu at the bottom of the sidebar: “User”. The “Profile” item in that same menu is where you choose your workspace (personal profile or organization).

Avatar. Click the avatar image to upload a photo. If no avatar is uploaded, the system shows a generated avatar with your initials. You can delete an uploaded avatar and go back to the generated one.

Basic details:

Name — your display name in the system. Other users see it in projects.

Nickname — a unique identifier used for search and mentions. Allowed characters: Latin letters, digits, hyphens, and underscores.

Job title — your position or specialization (optional).

About — a few words about yourself (optional).

Additional instructions — your directions for Iskra, up to 3,000 characters, for example: “keep answers short and to the point; I’m an engineer, technical detail is fine; format answers as lists.” They apply in your personal chats and are not visible to others.

Time zone — your time zone, chosen from the list. For details, see the “Time zone” section below.

To save your changes, click “Save”. The button becomes active only when there are unsaved changes.

Organizations and groups. The bottom of the profile page lists the organizations and groups you belong to, with your role in each.

Time zone

You set your time zone in Settings → Profile → Time zone: choose your zone from the list and click “Save”. A hint below the field shows your device’s time zone, which makes the right one easier to find if you don’t remember its name.

Iskra uses this time zone to understand “tomorrow at 9”, “in an hour”, and “at 21:00”: it sets task and reminder deadlines in your local time. The “Plans” section also shows entry times in this zone, the same on your phone, your work laptop, or while traveling.

Usually the time zone is already filled in: the platform takes it from your device settings at sign-up and at every sign-in. It is worth checking if you created your account on a machine whose clock is set to UTC, for example a work server.

Until a time zone is set, Iskra does not guess it: before setting a deadline in local time (“set it for 21:00”), it asks which time zone you are in and names that zone in its reply. “Remind me in ten minutes” works without a time zone, since local time is not needed there. However,

in the “Plans” section, times are counted in UTC until a time zone is set, so it is still worth setting one right away: depending on where you are, local time can differ from UTC by several hours.

You can change the time zone as many times as you like. The new zone applies to future entries and immediately changes how the times of existing entries are displayed. Entries created earlier still fire at the same moment: each one keeps the time zone selected when it was created (see “Plans”).

Showing your name in published chats

Show name and avatar in shared chats. This toggle controls what viewers of a published chat see. When it is on, your name and avatar are shown as the author. When it is off, your messages stay anonymous. See also Chat and conversations.

Changing your email and password

Changing your email. To change your email address, click the change email button. The process has two steps:

1. Enter the new email.

2. Confirm it with the 6-digit code sent to the new address.

Changing your password. To change your password, fill in three fields:

Current password — to confirm your identity.

New password — choose a new password. The strength indicator shows how strong it is.

Confirm new password — enter the new password again.

The “Change password” button becomes active only when all three fields are filled in correctly, the new password is strong enough, and both entries of the new password match.

Personalized replies: what Iskra remembers about you and how it knows your

preferences

Iskra tailors its replies to you, but it does not remember chats on its own. Everything it takes into account in a personal chat comes from three sources; when you ask “what do you know about me?”, it names the ones present in that chat without explaining how its reply is put together.

Saved at your request. A fact you asked Iskra to remember (“remember that…”) is saved by the built-in “Memory” skill if it is enabled (see “Skills”). A regular fact applies in all chats until you ask Iskra to forget it: Iskra lists such facts when you ask “what do you remember about me?” and shows them in memory.md in the artifacts panel. A note “for this chat” is a separate file, session-note-*.md, in the chat’s working folder and applies only in that chat. Pinned notes (/pin text; to list and remove them, use the “Pinned notes” window, opened by the button in the bottom row of the request panel or by the bare /pin command) apply until the end of a single chat.

Provided by settings and work context. Iskra receives your filled-in profile fields — name, organization, groups, role, job title, “About”, additional instructions — in every personal chat. Everything else is present only when it exists: the learning profile after the onboarding wizard, project instructions inside a project, Memory folders when you have connected them to the chat. Iskra does not mention a source that is absent. In a shared project chat, “About”, additional instructions, and the learning profile are not used; only the identifying fields remain: name, organization, groups, role, and job title.

Current chat. Everything you wrote and attached here, including “remember” requests that were not saved to the skill, Iskra takes into account until the end of the chat or until you click “Clear”, but none of it carries over to other chats by itself.

To make information apply in all chats, fill in your profile or ask Iskra to remember a fact. The “Memory” section in the panel holds your files and folders, not a list of facts about you.

Billing

The “Billing” (or “Subscription”) tab shows the active profile’s current plan, its parameters, and the history of changes, and lets you change your plan or return to the free one.

Pricing plans

Iskra has personal plans and separate package plans for organizations. The exact limit values and available features are shown on the pricing page and on the current plan card.

Personal plans apply only in your personal profile. They differ in MT limits, file storage, maximum file size, and available features, such as projects, schedules, or Memory.

Organization packages apply in the organization context and give all members a shared monthly MT pool. Organization membership is not limited by a number of seats, and the “recommended” hint helps you choose the package size.

Exact prices, MT limits, storage size, maximum file size, and the list of available features come from the current plan configuration. See them on the public /pricing page, on the current plan card, and, for organizations, on the Organization → Subscription tab (/org?tab=billing).

MT stands for megatoken, the unit of usage: 1 MT = 1,000,000 model tokens. Both your questions and the assistant’s replies are counted in the same currency. For more on usage, see the “Statistics” section below.

File storage is counted in total across the files you upload to chats and projects of the active profile. There is no separate limit on the number of files in a single chat or project, but a single file cannot exceed the available size limit.

The “Start” plan: a starter plan and a safety net

When you sign up, your personal profile automatically gets the Start plan. It is free and has no time limit: right after sign-up, you can chat with the agent within its limits.

Start also serves as a “cushion” when you leave a paid plan:

if you cancel your subscription, your profile returns to Start when the paid period ends;

if repeated automatic charge attempts fail in a row, the subscription is also downgraded to Start.

In both cases, work in chat does not stop; you simply continue on Start’s basic limits.

Subscription statuses

The subscription card shows the current status next to the plan name:

Active — the subscription is in effect and its limits apply. For paid plans, the dates of the current period are shown next to it.

Pending — a payment has been created, but the bank has not yet confirmed the charge. Complete the payment using the link in the banner or wait for confirmation; until then, your previous plan stays in effect.

Payment pending — the automatic charge did not go through on the first attempt. The platform gives you a 10-day grace period: you keep using your current plan while the system retries charging the linked card. If the payment still cannot be made within that time, the subscription is downgraded to Start.

Cancelled — you have scheduled a return to Start; the subscription stays in effect until the end date of the current period, after which it switches automatically.

Expired — the subscription has ended.

What the subscription card shows

The name of the current plan and its status.

Monthly price: for a personal plan, the plan price; for an organization, the price of the selected package.

Validity period: the start and end dates of the current cycle. For the Start plan, it says “Lifetime”.

The list of included MT limits, chat file limits, and the list of available features. Organization packages show the organization’s shared monthly pool.

The monthly MT limit applies within the paid month and is counted from the purchase date, not from the start of the calendar month. The window moves to the next month when a new paid period begins; usage within the current window is kept. If you renew your subscription early, the limit is not reset at the moment of payment; the new count starts when the new period begins.

The “Plan change scheduled” banner appears if you have chosen a different plan starting with the next cycle.

The “Subscription cancelled” banner appears if you have returned to Start; it shows the date until which your current plan remains in effect.

The “Payment not completed” banner appears if the last payment has not been confirmed yet; its buttons let you complete the payment or cancel the request.

Changing your plan

The “Change plan” button on the personal subscription card or the “Pricing” link in the navigation takes you to the page for choosing a personal plan. It shows how many times the monthly limit increases compared with Start and which features are included. For an organization, the owner chooses a package on the Organization → Subscription tab (/org?tab=billing), which lists package plans with their monthly MT pool, recommended number of users, and storage.

What happens when you choose a plan depends on how the prices compare:

Upgrade — a confirmation window opens. If you already have a paid plan, its title says so: “Plan change: current → new”, since this is a switch, not a first subscription. The window shows how much will be charged now, how much is credited for the remainder of your current plan, and the date and amount of the next scheduled charge. After you confirm, the system redirects you to the YooKassa payment page. Once the payment succeeds, the new plan takes effect immediately: the billing period and the monthly token usage window start over from the moment of payment. The credit is calculated from the amount actually paid for the current period, including any discount applied when you bought it. The smaller of two shares is used, remaining days or remaining MT:

for example, with 70% of the days and 20% of the MT remaining, 20% of the payment is credited. A period granted for free gives no credit. A promo code for the new plan is applied first, then the credit is deducted; the credit cannot exceed the price of the new plan. If the next period has already been paid in advance, you can upgrade after it begins. If MT usage is temporarily unavailable, try the calculation again later.

Downgrade — a confirmation window appears. The new plan takes effect from the next billing period; until then, you keep your current limits. Before the switch date, you can cancel the change with the “Keep current plan” button.

A switch is already scheduled — if you choose the same plan again, the system shows a reminder and does not create a second payment.

You can stop renewal with the “Auto-renew” toggle on the subscription card. Before it turns off, we ask what didn’t work for you: you pick a reason from a short list, and both the comment and the answer itself are optional — you can skip this step, and it has no effect on turning off renewal. The same window shows the date until which you keep access. Your current paid plan keeps working until the end of the paid period, after which Start is activated. The decision is reversible: until the period ends, you can turn auto-renewal back on with the same toggle. Refunds for the unused remainder of the period are not provided.

Resetting the monthly limit

If your paid monthly limit runs out before the end of the period, plans from “Pro” and up offer a way to get back to work immediately, without waiting for the window to refresh automatically. The “Reset limit” button appears on the subscription card on the “Billing” tab as soon as you have used at least 90% of the current period’s monthly limit, and on the limit-reached card in chat the same option is shown as a separate line below the offer to change your plan (on the Ultra plan, which has no upgrade, the reset is the main action).

A reset is a payment for your current plan at full price, without active discounts and without credit for unused days: it is a one-time purchase, not a subscription renewal. After payment, the plan’s tokens are granted again in full, the billing period restarts from the moment of payment, and the next charge moves one month ahead of that date; the interface shows the exact date in advance in the confirmation window. Unused days of the previous period are forfeited: the confirmation window tells you how many before you confirm the payment.

A reset is not the same as a renewal: a renewal pays for the next period in advance and does not affect current usage, while a reset pays for the current moment and zeroes out usage right away. If a switch to another plan or cancellation of auto-renewal is already scheduled for the subscription, the reset does not cancel it; that decision still applies at the end of the new period.

A reset is unavailable if the subscription has already been renewed in advance and the paid period has not started yet; in that case, the interface tells you that the reset becomes available when the already paid period begins. A reset you have already paid for is not refunded automatically; if you need a refund, contact support.

Payment via YooKassa

Payments in Iskra go through the YooKassa service. On the payment page, you can choose a convenient method: a bank card or SBP (Faster Payments System). The payment method you used to pay for the subscription is linked to it automatically and used for subsequent automatic charges.

After payment, Iskra opens a confirmation screen showing what was paid, until what date the subscription is valid, and when the next charge will take place. If the payment was not completed, the screen says so and offers to take you back to plan selection. Possible reasons for a decline: insufficient funds, an expired card, the bank declined the transaction, 3D-Secure authentication was not passed, or the issuing bank is temporarily unavailable. In these cases, you can try again or choose another payment method.

A personal paid plan is a subscription with auto-renewal: at the end of each paid period, Iskra charges the saved payment method and extends your access. About a day before each charge, we send you a reminder email. Auto-renewal is turned off with the toggle on the subscription card: we ask for the reason (answering is optional), and access remains until the end of the paid period. The saved payment method stays in place, so turning renewal back on takes just one click, with no questions.

The saved payment method is shown in the “Payment method” block. For a card, you see the last four digits and the expiry date; for SBP, a mark that it is linked. The “Manage payment method” action reveals the removal option: first you tick a confirmation checkbox, and then the same window asking for the reason opens. Removing the method turns off auto-renewal, so if you simply want to stop paying, use the auto-renewal toggle instead, since it is reversible. If the payment method has been removed while the subscription is still active, the card shows a reminder with a “Renew and save card” button: paid days are not lost with this renewal — the new period is added to the current one, the payment method is saved again, and auto-renewal turns back on, so from then on the subscription renews by itself. If a switch to another plan is scheduled for the subscription, this renewal is charged at your current plan, and the switch moves to the end of the new period; the reminder tells you this before you pay.

An organization package is paid for as a monthly package. When paying, the owner can turn on auto-renewal; if auto-renewal is off, the package can be renewed or resumed by card or by invoice.

Changing the package. The owner can switch to another package in the “Change package” window. When moving to a larger package, the window shows the amount due, the credit, and the next charge date in advance, before payment. The credit is the unused part of the paid current period: the earlier the switch and the less the organization has spent from its monthly pool, the larger the credit. If the current period was received for free — via a promo code or from an administrator — there is no credit, and you will need to pay the full price of the new package. When the credit covers the full cost of the switch, the package changes immediately, without payment; otherwise, you can pay by card or by invoice. After payment, the pool immediately grows to the new size and starts counting anew; the paid term is not shortened, and the next charge date moves forward. If you recently renewed the subscription in advance and the paid period has not started yet, changing the package becomes available once that period begins. Moving to a smaller package requires no payment: it applies from the next period, and until then the current package stays in effect.

If the organization has been assigned a period with a non-standard term, such as an annual contract, you cannot change the package yourself: such a switch is arranged through support.

If the shared pool runs out before the end of the period, the owner can buy additional megatokens in the “Buy tokens” window. The price of one additional MT depends on the organization’s package: the larger the package, the cheaper the top-up. The current price and the total amount are shown in this window before payment, both when paying by card and when issuing an invoice. If the organization has a permanent discount, the total and the price of the top-up package are shown with the discount applied, with the full amount struck through next to them; the discount does not reduce the number of megatokens. If the price cannot be calculated for your package, the window says so and the payment buttons are unavailable; in that case, contact support and we will help you arrange the top-up.

If a top-up paid by card is refunded, the megatokens credited for it are revoked along with the money, to the extent that they have not been used by that time. With a partial refund, the corresponding part is revoked. If the top-up was paid by invoice, the refund is arranged through support: write to us and we will explain the procedure.

Organization subscription

If you create your own organization (see Account, profiles and organizations), the organization itself is created free of charge. However, work operations in it — shared spaces, projects, schedules, Memory, chat with the agent on behalf of the organization — require the organization’s own package.

Until the organization has a package, any such operation returns an “Organization not active” notice with a button that takes you to payment. The owner sets up the package on the Organization → Subscription tab (/org?tab=billing), and it covers all of the organization’s members. Members’ usage is deducted from the organization’s shared monthly pool; the organization administrator can additionally assign personal quotas to members.

Paying by invoice (for legal entities)

An invoice for buying or renewing a package takes the organization’s percentage promo code into account. The amount is fixed when the invoice is issued and stays the same until payment. If the code was applied after an invoice had been issued at full price, cancel that invoice and issue a new one. The promo code period is used up only after the payment is confirmed.

If it is more convenient for the organization to pay by bank transfer rather than by card, the owner can issue an invoice right on the billing tab, without contacting support. The “Pay by invoice” button is available next to card payment in five places: when activating a package, when renewing or resuming a subscription, when moving to a larger package, when buying additional tokens, and when buying an additional top-up package.

Company details. When you issue an invoice for the first time, you need to fill in the payer’s details. You do this only once; after that, they are used for all subsequent invoices:

INN (taxpayer ID) — as you type it, the system tries to fill in the organization name, KPP, and legal address automatically (if the EGRUL (Unified State Register of Legal Entities) database is connected on the platform). If the company is not found or autofill is not configured, fill in the fields manually.

Legal entity name and legal address.

Accounting email (optional) — you can send a link to the invoice to this address.

You can change the saved details at any time with the “Edit” button; this does not affect an invoice that has already been issued: its amount and details are fixed at the moment of issue.

Issuing the invoice. The next step shows what you are buying and the total amount. The “Issue invoice” button generates a PDF instantly and assigns the invoice a number in the form “ИО-101”; give this same number in the payment purpose. The invoice is valid for 14 days, except for an invoice for moving to a larger package: it locks in the credit for the remainder in advance, so its validity is shorter, 5 days.

Until the invoice is paid or cancelled, its card is shown on the organization’s billing tab with the number, the validity period, and these buttons:

“Download PDF” — get the invoice file to pay it through your bank’s client system;

“Send by email” — queue an email with a link to the invoice for sending to the accounting email specified in the company details. The email goes out within a few

seconds, and its status (“pending”, “sent”, “delivered”, “delivery failed”) is shown in a line below the button;

“Cancel invoice” — if you no longer need the invoice (for example, the details were entered incorrectly or you changed your mind).

An organization can have only one active invoice at a time. If an invoice has already been issued and you try to arrange another payment by invoice, the system offers to pay or cancel the existing one before issuing a new one.

Activation after payment. Invoice payments are checked manually: as soon as the payment reaches our bank account, our team marks the invoice as paid, and the package is activated automatically. This usually takes 1–2 business days after the funds arrive. While you wait for the funds to arrive, nothing stops you from changing your mind and paying by card through the same action: the invoice does not block anything, you just need to cancel it.

If auto-renewal is not enabled for the package being renewed, we recommend not putting off payment by invoice until the last day: a bank transfer usually takes 1–2 business days, so it is better to issue the invoice in advance. We will remind you about this by email about a week before the period ends.

Learning profile

This tab is available only when the learning feature is enabled on the platform. It is hidden by default.

The learning profile contains information collected during onboarding and helps Iskra choose hints and examples that better fit your tasks.

If you have completed onboarding, the page shows your data:

Roles — your role or specialty (for example, developer, analyst, manager).

Experience — your level of experience with AI.

Work tasks — the typical tasks you handle.

Learning goals — what you want to learn (if you specified them).

You can go through onboarding again by clicking “Redo Onboarding”. This is useful if your tasks have changed.

If you have not completed onboarding yet, the page offers to start it.

Audit

The “Audit” tab contains a log of security violations detected by the system (for example, attempts to override the assistant’s instructions).

Entries are shown in a table with the following columns:

Date — when the violation was recorded.

Status — the severity level of the incident.

Category — the type of violation.

Source — where the request came from.

Message text — the content that triggered the detection.

Statistics

The statistics page shows information about your use of the platform in the active profile.

A quick way to get here is the “Usage” item in the profile menu at the bottom of the sidebar: on its right, it shows the percentage of the monthly limit used, the same as in the indicators on this

page. The item is present only for profiles with a monthly limit.

Current limits. The top of the page shows usage indicators in megatokens (MT). This is the unit of usage; the base ratio is 1 MT = 1,000,000 tokens, and the final value may include the mode multiplier shown in chat:

usage for the current subscription month;

the monthly limit of your current plan and the date when the window refreshes;

a progress bar.

MT usage values already reflect the actual mode of each step. Iskra does not raise the power level on its own: you choose the maximum level, “Very smart”, in the chat yourself. While it is selected, the “Complexity” button shows a “×N” mark: N is the multiplier of the model that will actually perform the step, and it is this multiplier that applies to your charges. If your organization uses its own set of models and it has no higher-tier model, the maximum level remains available but is shown as ×1: the reply comes from the regular model at the regular price. If a failure occurs during a reply and Iskra continues on a fallback model, subsequent steps are counted at that model’s multiplier; after a return to the regular level, they are counted by that level’s model again.

Internal model calls — when Iskra assesses the complexity of a request and selects suitable skills — are not billed to you: they are free and do not count toward your monthly usage.

The chart of input and output tokens shows the actual volume of text; to compare against your limits, go by the MT values.

If the limit is exhausted, sending new messages in chat is blocked until the window refreshes, which happens when a new paid period begins. You can upgrade to a higher plan right away, without waiting for that date. The parameters of your plan (plan, limits, features) are shown on the “Billing” tab.

30-day chart. A line chart shows your token consumption over the last 30 days. The chart has two lines:

input tokens (your messages and context);

output tokens (the assistant’s replies).

Hovering over a point on the chart shows a tooltip with the exact values for that day.

Summary statistics:

total number of tokens for the month (input and output);

average number of tokens per day.

Top chats. Two lists of the most “expensive” chats by token consumption:

top chats of all time;

top chats today.

Each entry shows the chat title, the number of input and output tokens, and the total usage. Click the title to go to the chat.

Storage

The “Storage” tab shows how much file space is used and helps you tidy up your uploaded files. In a personal profile, this is your quota. In an organization profile, the organization’s total capacity is shown at the top, and the list below contains only your chats and the projects you own: colleagues’ files are not listed and are counted as a single line, “Other members’ files and service files”.

Total usage. At the top, you see how much space is used relative to your plan’s limit and how full it is as a percentage.

Files by section. Uploaded files are grouped by source: personal Memory, projects, chats, and synced folders. You can collapse groups, filter the list by type, and search it by name.

File actions. You can open a file in preview or delete it. Deleting frees up storage space.

If you are running out of space, you can go from here to changing your plan.

Public links

The “Public links” tab collects all chats of the active profile that you have shared via a link and lets you close access when a link is no longer needed. The section is always available: even if creating new links is temporarily disabled on the platform, you can still revoke the ones you created earlier.

The section has two tabs: “Chats”, with the chat links described below, and “Boards”, with the pages where you show files to readers.

Counter — how many public links are currently active in the active profile.

List row — the chat title, the publication date, and the date the link’s content was last updated.

Open chat — go to the original chat. If the chat was archived and then deleted from the archive, there is no way to open it, and a “conversation unavailable” mark appears next to it; you can still close the link itself from here.

Open public page — see what readers see.

Copy link — copies the address to the clipboard.

Close access — revokes a single link, with confirmation. After that, the page at this address no longer opens, and attached files become unavailable.

Close access to all links — revokes all links at once; the confirmation window shows how many links will be closed.

A row may show a “the link does not open right now” mark. This is the server’s live answer about serving the link at the time of the request, not something stored in the row: it marks a link that has been blocked by the organization’s external access policy or that is waiting for administrator approval. The link itself has not been revoked, and you can still close it from here.

Revocation is irreversible: the previous address cannot be restored. The chat itself, its messages, and its files are kept; only public access is closed. If you share the chat again later, Iskra creates a new link with a different address. For more on creating and updating links, see Chat and conversations.

For an organization profile, links are also closed automatically when a member is removed, leaves on their own, or when the organization is deleted; you do not need to revoke them manually after that.

The list is tied to the active profile: switch profiles in the sidebar to see the organization’s links.

The “Boards” tab

The “Boards” tab collects the boards of the active profile that you manage, including restricted ones: “Only me”, organization boards, and group boards. In chat, you see only the boards of the current conversation, so the full list lives here.

The tabs of this section are independent of each other: closing a public chat link does not remove a placement from a board, and deleting a board does not close chat links. Both belong to the active profile; to see the organization’s boards, switch profiles in the sidebar.

A board’s contents, its audience, version pinning, deletion, and restoration are described in the “Boards” section.

Antonym — private mode

Antonym is a privacy mode. When it is on, your requests are processed by a language model on Iskra’s servers and are not sent to an external model provider. The request text itself does not change: what changes is where it is processed, not its content. That is why the mode is suitable for personal data, trade secrets, and documents marked for official use only. This applies specifically to the external model provider: the list below names the surfaces the mode does not cover.

What the mode covers: the text of your request, the agent’s work (including tool results and intermediate steps), and the model’s text reply.

What stays outside the mode — these still take the platform’s usual route:

voice input (speech recognition);

spoken playback of the model’s reply — playback is currently not enabled anywhere, so in practice this path is not used; the limitation will apply once it returns;

the scheduled “task is done” email, if Email notification is enabled for the task: the email body contains the full text of the result;

submitting a skill for moderation;

the summary for the platform administrator.

Mobile push notifications do not belong to this list: in private mode, they arrive without a preview of the reply text.

The Antonym toggle is available in your personal profile settings. For organization profiles, the setting is managed at the organization level by its owner or administrator.

How to turn it on: open settings → the “Security” tab → the “Antonym — private mode” toggle. The change takes effect within about 1 minute. For an individual chat, you switch the mode in the bottom row of the request panel (the shield button).

When the toggle is unavailable:

The mode is not open on the platform — the toggle is hidden or locked if the administrator has not opened private mode for your organization.

Enforced mode — if the administrator has made the mode mandatory, Antonym works for everyone and the toggle is locked. You cannot turn it off in an individual chat either.

How private mode differs in practice:

a different model answers, so replies to complex tasks may differ, and the complexity levels collapse into one;

the mode has no fallback model: if Iskra’s servers are busy, the request is not executed and a message with a “Retry” button appears. Since the request was not executed, you are not charged for it;

if a required capability is not available on Iskra’s servers (for example, recognizing scans), Iskra says so directly: retrying will not help here, and the issue is for the administrator to resolve;

switching the mode while a reply is in progress takes effect from the next message.

Chat indicator: when the mode is active, the chat interface shows a “Private mode is on” label — below the input field in a new chat and as a compact badge above the field in an active one. Clicking it opens a window that explains what the mode covers, what stays outside it, and where to manage it.

A detailed technical description of Antonym is available in the administrator documentation.

Referral program

The referral dashboard lets you invite new users to Iskra and earn rewards for it. The tab is available if the referral program is enabled on the platform.

Your referral link — your personal invitation link; you can copy it or send it as a ready-made banner image. A user invited through it gets a discount on their first payment (if one is configured).

Banner appearance — a switch above the image: “Match interface”, “Light”, or “Dark”. By default, the banner follows the app theme and changes along with it; the light and dark variants fix the look regardless of the theme, which is handy when the image goes into a chat or a presentation with a different background. The choice is saved in the browser and persists after the page reloads. It does not affect the banner’s content (the link, the discount size); copying and downloading the PNG give you the selected variant.

Your terms — a block with your current terms: the percentage you earn from invitees’ personal subscriptions, the percentage from their organizations’ payments, how many payments of each invitee count, the minimum payout amount, and the discount invitees get. The wording “first N payments” means the first N payments of each invitee made after the program launched; personal and organization payments are counted together. “All payments” means there is no limit. The invitee discount applies to their first personal subscription payment; if there is no discount, the line is not shown. The platform may change the terms; the block then shows the new values, and rewards already accrued stay the same.

Statistics — how many people came through your link and how many of them subscribed, as well as the revenue from your referrals.

Rewards — how much is available for payout, how much is still awaiting confirmation, how much is being processed, and how much has already been paid out, as well as the minimum amount and the date of the next payout.

Payout request — once you have accumulated the minimum, you can submit a payout request by accepting the terms of the offer. The reward amount is taken from the “Your terms” block.

Application

General application settings and technical information.

Theme. Three options: “System”, “Light”, and “Dark”. With “System”, Iskra follows your operating system’s current theme and switches automatically when the system theme changes. Light and Dark fix the selected mode. The theme applies to the entire main interface: sidebars, chat, Memory, projects, schedule, skills, and help.

Interface language: Russian, English, French, Chinese

Choose Russian, English, French, or Simplified Chinese. The interface switches immediately; your choice is saved in this browser. Messages and documents already written are not translated automatically.

Response language

Sets the default reply language in the web chat separately. In “Follow interface language” mode, Iskra uses the language selected in the current window. You can also explicitly choose one of the four languages; then changing the interface language does not change the response language. To link them again, choose “Follow interface language”.

The response language setting is saved automatically for your account, applies in all of its profiles, and is restored after you sign in again. Another account uses its own setting. A new value applies from the next message you send or the next regeneration; a reply already in progress continues with the previous setting.

You can ask in a message, “reply in English”: the request applies to that reply and does not change the saved choice. The language of a message by itself, without such a request, does not switch the default reply language. Applicable additional instructions of the profile and the project are still taken into account. The setting applies to regular personal and project web chats; some service replies, schedules, and integrations follow their own rules.

Delete all chats. A button that deletes the entire chat history of the current profile. The system asks for confirmation first. This operation is irreversible: all chats, messages, and related data are deleted. All public links of this profile are closed along with the chats: pages at the old addresses stop opening, and they cannot be restored.

Next message suggestions. When this feature is enabled on your platform, you can turn off or bring back the suggestions shown in the empty chat field here. The choice is saved in this browser. You can send a suggestion right away with Enter or move it into the field for editing with Tab, the right arrow key, or a tap. For details, see Chat and conversations.

Build information. The bottom of the page shows technical information: the product version, the commit hash, the build date, and the database schema version. This information may come in handy when you contact support.

Overview

Iskra has three types of promo codes:

1. Free plan — gives you a subscription for N days without payment.

2. Discount — gives you X% off the next N payments.

3. Usage reset — zeroes out the tokens already used in the current month.

All of them are activated in the plan settings: Settings → Plan & billing → Activate promocode. The code field is always visible, even if you already have an active paid subscription.

This section covers personal promo codes. Promo codes for an organization subscription (bonus MT, a free period, a discount) are activated by the organization owner; they are described in the Organization section.

Overview

Enter the code and click “Activate”. If the promo code is valid, you immediately get a subscription to the specified plan (Pro, Expert, or another) for N days. When the term ends, the subscription automatically switches to the “Start” plan.

You can activate it only when you have no active paid subscription. If you already have a paid one, wait until it ends.

Overview

Discount promo codes are linked to your account and apply to your next payment, whether it is the purchase of a new plan or the auto-renewal of your current one.

Cases:

If you don’t have a paid subscription yet, enter the code, then choose a plan and click “Subscribe”: the price is recalculated with the discount automatically.

If you have an active subscription, the discount applies at the next charge (renewal).

The discount applies to a limited number of payments (for example, 3 periods × 25% off). Once they are used up, the subscription renews at full price.

Plan restrictions. A discount can be tied to specific plans (for example, only Pro and Expert). The active discount page in settings shows which plans it applies to.

Overview

Sometimes a discount appears without a promo code. If you run out of your monthly token limit on the “Basic” plan, Iskra itself offers you a switch to “Pro” at half price: the “Out of tokens” block shows the discounted price, the struck-through full price, and the date until which the offer is valid.

What to know about this offer:

you don’t need to enter anything: the discount is already linked to your account;

it applies only to switching to “Pro” and only to the first month; after that, the subscription renews at full price;

the offer is available for a limited time (usually a week); if you don’t use it, it expires;

only one discount can be active at a time. If you already have a discount from a promo code or an invitation, no new one is issued, and the one you have stays in effect;

next month, if the limit runs out again, the offer may appear again.

The “Out of tokens” email is about the same offer and states the same price: the discounted one, with the struck-through full price next to it, and the date until which the offer is valid.

Overview

When you have an active discount promo code, the Settings → Plan & billing page shows an “Active discount” card with:

The promo code and the discount percentage (for a discount without a promo code, the card says where it came from instead of showing a code, for example “Discount for a used-up limit”)

Progress: “N of M periods remaining”

Validity (the date until which it can be used)

Applicable plans

Overview

The “Active discount” card has a Cancel button. After you confirm, the promo code is deactivated, the remaining discount periods are forfeited, and you can enter a new promo code.

Important: a cancelled promo code cannot be reactivated on the same account; this is a one-time action. Use it when you want to enter a more favorable code instead of the current one.

Overview

This type of code zeroes out the tokens already used in the current month, as if the usage window had just started. Your next charge date and current plan do not change: this is neither a reset purchase nor a subscription renewal, but a one-time write-off of your usage.

You can activate it even with an active paid subscription: the promo code field is always available precisely for codes like this.

When it pays to activate. The closer your usage is to the monthly limit, the more useful the reset. If less than half of the limit has been used when you enter the code, Iskra asks you to confirm: “You’ve used X% of your limit — resetting pays off more the closer you are to it. Activate now?” You can confirm, or cancel and keep the code for later, activating it when your usage is closer to the limit.

If the code does not work:

“This code only works in a personal profile” — a usage reset is not available in an organization profile. Switch to your personal profile and enter the code there.

“Finish or cancel the limit reset purchase you already started” — if you have an unfinished paid limit reset purchase, first pay for it or cancel it, and then activate the promo code.

Like other promo codes, each usage reset code can be used once per account. You can activate another code of this kind later in the same month without restrictions.

Overview

“Code not found” — check the spelling.

“Promocode expired” — the promo code’s validity period has ended.

“You already have an active discount promocode” — cancel the current one in settings to enter a new one.

“You have already activated this promocode” — each code can be used once per account.

“You already have an active subscription” — applies only to free plan promo codes; wait until your current subscription ends.

“This code only works in a personal profile” — applies to a usage reset promo code: switch from the organization profile to your personal one.

“This profile has no personal subscription” — applies to a usage reset promo code; this almost never happens.

“Finish or cancel the limit reset purchase you already started” — applies to a usage reset promo code: the subscription has an unfinished paid limit reset purchase.

“Too many attempts” — you have exceeded the attempt limit (5 per minute). Wait a minute.

Overview

Skills extend what the assistant can do: they are additional tools and actions it can perform when a plain text reply is not enough. For example, checking the current weather, converting units, looking something up in Wikipedia, calling an external service, or applying your own instruction.

Usually it is enough to write your request in chat, and Iskra picks a suitable skill itself. In the Skills section (/skills), you can choose an operating mode for each skill. A skill that can be called explicitly can be added with the “/ Skills” button on the start screen or with the “Add a skill” button in the “Conversation skills” window of the request panel in an ongoing chat, or you can find it by typing / and part of its name or description in the message field; the search understands the skill’s Russian name and words describing its purpose.

Overview

The Skills section opens from the “Skills” navigation item or the direct link /skills. The top level separates two tasks: “My skills” opens by default, and “Catalog” (if it is enabled on the platform) is for finding and installing new skills. To the right of the section heading are a compact search button (it expands into a field when clicked) and the “Run history” button; the “+ Add your own skill” button sits to the right of the status filter, above the list.

“My skills” and the status filter

All your skills are shown as cards, two per row, grouped into sections by ownership:

Personal — skills you created or submitted yourself. While you have no such skills, the section is not shown at all; you can create your own with the “+ Add your own skill” button above the list.

Organization — skills published in your organization and catalog installations made for the organization. The section is visible only when you work in the organization context.

System — built-in skills available to everyone out of the box (see below) and internal system skills such as “Task advisor”.

From catalog — skills you have installed for yourself from the catalog showcase.

Above the sections is the status filter, a button that shows the current value and opens a list when clicked: “All”, “Enabled”, “Disabled”, and “In moderation”. It changes only which cards appear on the “My skills” page: “Enabled” shows skills in “Auto” and “On command” modes, “Disabled” shows turned-off skills, and “In moderation” shows submissions that are not yet published. “All” is selected by default. Each section shows all matching cards at once.

Search (the magnifier button to the right of the section heading) looks through names, technical names, and descriptions among the cards left after filtering; on the “Catalog” tab, the same search looks through the showcase.

A card shows the skill’s readable name, the technical name for calling it via “/” (when it differs from the name), and a short description. For built-in skills, the name and description are shown in the interface language; for skills by other authors, the text stays as the author wrote it, and in a non-Russian interface the description is preceded by an “author’s text” mark (for details, see the “Showcase” section in the skill catalog help). A disabled skill shows “Off” in the top right corner of its card, after the action icons if there are any; enabled skills get no separate mark. The skill’s icon is shown on the left if it has one; if it has none or the icon fails to load, a colored placeholder with the first letter of the name is shown instead. Clicking the card itself opens the skill page; there is no longer a separate “Details” button. The top right still holds

icon buttons for actions. If an installed skill is missing required keys or parameters, a warning button appears; its tooltip reads “Needs setup”. Click it to open the required form right away. Once the required values are filled in, the button disappears. When all the missing values are set by an administrator rather than by you, there is no form: on the skill page, the line “Managed by the administrator” appears instead of the setup button. If an already published skill has an update submission, an “Open my submissions” icon appears next to it; it leads to the detailed moderation status, while clicking anywhere else still opens the skill itself.

Details page

The “← Back” link returns you to the skill list. The header shows the name and, if it differs, the technical name for calling the skill via “/”. In the top right corner is the Off / On command / Auto mode switch with a short caption explaining the selected mode; system skills that cannot be turned off show an “Always active” pill instead of the switch.

Below the header are horizontal tabs: Overview, Tools (with the number of tools), Settings, and others if they are available for the skill. On “Overview”, the skill’s version and its ownership (Personal / Organization / System) appear above the description, along with the moderation status if the submission is still under review. Next to the update status of a published skill, there is an “Open my submissions” link.

At the bottom of “Overview” is the “Skill management” section. It shows only the actions available to you in the current profile: publishing or updating in the catalog, updating the source skill, transferring to an organization, and deleting. Deletion is set apart from the other actions and always requires confirmation; if you do not have the required permission, the corresponding button is not shown.

If a skill or its new version is still under review, the card shows a compact status: “Under review”, “Approved, publishing”, “Rejected”, or “Build failed”, always with the version next to it. For a new submission that has not been published yet, the card is inactive, with no operating status on the right. Click “Details” to open “My submissions” for the relevant profile and see the detailed stage or the reason for rejection. For an update to an already published skill, the previous version keeps working until the new one is successfully published. To open the status of such an update, click the “Open my submissions” icon on the card or the link next to the status on the skill page; “Details” still opens the skill itself.

While a submission is being processed, the catalog updates its status automatically. If statuses temporarily fail to load, the skills themselves remain available without them: you can click “Retry” in the warning. Cancelled submissions are shown only in the “My submissions” section.

The “Manage skills from chat” hint

If managing skills from chat is enabled on your platform, a hint window appears when you open the section. It reminds you that the easiest way to manage skills is simply to ask Iskra in a chat, and it lists what Iskra can do: create a skill from a plain-language description; enable, disable, update, or delete one; set parameters and keys in a protected form; find and install a skill from the catalog. The “Examples” tab collects ready-made phrases: click any of them, and a chat opens with the text already filled in; all that remains is to complete the highlighted part and send it. The familiar way, using the buttons and cards of this section, of course still works too.

When you create a text skill in chat or update one (in chat or in the Skills section), describe the capability you need in plain words. If the instruction contains another skill’s internal tool name in the form skill__skill-name__tool-name, Iskra stops the creation or update and asks you to remove that reference. Describe

the purpose of the capability you need in words, and state that a suitable capability should be found while the skill is running.

Both tabs, “Features” and “Examples”, consist of two side-by-side sections: “Skills” on the left (described above) and “Scenarios” on the right. The “Scenarios” section briefly explains that a scenario saves steps on top of a skill, how to create one through a rehearsal and run it with a phrase or a chip from the slash menu, and that a finished scenario can be published to the organization; on the “Examples” tab, the right side offers ready-made phrases of the same kind for scenarios (for details, see “Skill scenarios” below). The “Scenarios” section is shown only if this feature is enabled on the platform; otherwise, the “Skills” section takes up the whole window.

The window appears every time you open the section until you tick “Don’t show again”; the choice is remembered in the current browser, so the hint appears again on another device. To close the window without ticking the box, click “Got it”, click the close (×) button, or press Esc.

Operating modes

You change the mode on the skill details page (the switch in the upper right corner), and it takes effect immediately:

Off — the skill is not added automatically, does not run in the background and is not shown in explicit selection lists.

On command — Iskra does not add the skill on its own; you can call it only by explicit selection: with the “/ Skills” button, “Add skill” in the “Conversation skills” window, or the slash menu.

Auto — Iskra adds the skill on its own when it fits the request; explicit selection in the same ways and background work are also available.

Background skills offer only “Off” and “Auto”: background work cannot be enabled for manual calls only. Sometimes some switch positions are unavailable because of an administrator policy.

You can switch a skill to “On command” or “Auto” right away, even if it is not configured yet — you don’t have to fill in keys and parameters. If the skill needs your consent to write to secrets, Iskra shows a consent dialog before applying the mode. Switching to “Off” requires no setup.

A skill that lacks the setup it needs to work is still visible in the “/ Skills” list and the slash menu, marked “not configured”: it is not hidden from the list — when you call it, Iskra explains which data is needed and offers to fill it in.

An explicit call stays visible after the message is sent: the message text keeps /name, and the skill chip appears below the message. A copied message carries the call over and pins the skill when you paste it into the input field, while deleting the / token when editing the message removes the skill (for details, see “Choosing a skill or scenario” in the chat help).

Action confirmation

Some skills show a ready draft before acting and wait for your confirmation: email — a message with the recipient, subject and text; Max messenger and Telegram — a message with the exact chat; calendar — the assembled meeting; Google Drive — an edit or deletion of a file; CRM — a ready deal or task; email campaigns — the prepared campaign. If you confirm the same thing every time, you can remove this step.

On the skill page, open the Settings tab — it has an “Action confirmation” block with two options:

Ask every time — the way it has always worked: the skill shows a draft and waits for you to click “Send”. This is the default.

Allow and don’t ask again — the skill performs the action right away, and no form appears. The result shows up in the chat as a regular skill message (“Email sent” or an error), and the tool call itself is kept in the conversation as always.

The choice applies only to your profile and only to this skill. It also covers scheduled runs: the skill will send without you.

The choice does not affect the skill’s questions — the setup wizard, account connection, chat linking and clarification forms still wait for your answer. Nor does it affect forms that exist so you can edit something or for your protection: for example, confirmations in the skill-management skill are always shown. If a draft turns out to be incomplete (for example, the email body is empty), the skill still shows the form so you can complete it.

Not every skill has this block: the platform administrator can turn this option off for a specific skill.

Run history

The “Run history” button above the tabs opens the log — a history of your skill calls with their status (success, error, timeout, cancellation), time and which skill was called. You can filter the list by status and by a specific skill; clicking a row expands the run details. The “← Back” link returns you to the skill list.

Manual run

Besides regular chat, some skills can process documents or tasks one at a time, in a persistent chat, at the click of a button. On the details page of such a skill, the “Overview” tab has a “Manual run” panel: the queue size, the time of the last run and the “Run” button.

Clicking “Run” opens the skill’s persistent chat (repeat runs use the same chat) and starts processing the next item in the queue — you don’t need to write a request yourself.

Overview

Iskra’s user catalog usually includes built-in skills that are ready to use without setup. They work right away: there is nothing to configure and no access keys are needed. These skills get the “Auto” mode by default — if you don’t need one of them and the platform policy allows changing the mode, find it in the “System” section and select “Off” on its details page.

Some system skills call paid external services and require a corporate key set by the platform administrator. Users don’t need to enter such a key; until the administrator configures it, the skill is unavailable.

Weather

Current weather and a 1–7 day forecast by city name.

Example requests:

“what’s the weather in London”

“5-day forecast for St. Petersburg”

“will it rain in Berlin tomorrow”

Currencies

Exchange rates from the Bank of Russia and conversion of amounts between currencies.

Example requests:

“yuan exchange rate”

“convert 100 dollars to euros”

“how much is the dollar right now”

Dates

The difference between dates in days, shifting a date N days forward or back, and counting working days, weekends and public holidays in a range according to the official Russian working-day calendar.

Example requests:

“how many working days are left until the end of the month”

“what day of the week will it be in 90 days”

“how many days are there between January 1 and May 9”

Wikipedia

Search for Wikipedia articles and a brief summary by article title. Russian and English are supported.

Example requests:

“find Quaternion on Wikipedia”

“a brief summary of the Photosynthesis article”

“what does Wikipedia say about Turing”

Units of measurement

A unit converter: length, mass, temperature, volume, area, speed, data size and time. Iskra detects the category automatically.

Example requests:

“convert 5 km to miles”

“what is 100 degrees Fahrenheit in Celsius”

“100 megabytes in gigabytes”

Working with PDFs

The /pdf skill reads, creates and edits PDFs: it extracts text, tables and images, merges and splits pages, rotates and crops them, fills in interactive and flat forms, recognizes scanned text in Russian and English, adds watermarks and protects the file with a password. Just attach the document and describe the result you want; no separate keys or settings are needed.

Iskra keeps the original file and creates a separate result by default. Before handing it over, it reopens the finished PDF and reviews the relevant pages as images. Regular page review does not add PNG files to Drafts; images are saved separately when they are needed as a result. For a table, you can ask for CSV or JSON, and if you need XLSX, say so explicitly in your request.

In a flat form, added text becomes regular page text: no note appears when you hover over it. The original comments and genuine interactive form fields are preserved.

Example requests:

“extract the table from this PDF and save it as XLSX, don’t lose the leading zeros”

“fill in the form fields and save a new PDF”

“recognize the Russian and English text in this scan”

“merge these PDFs, set a password and check the final file”

A complex PDF may contain non-standard fonts, damaged objects or a layout that cannot be reproduced exactly in an editable document. In that case, Iskra reports what it was able to verify and does not promise an exact PDF→Word conversion or preservation of an arbitrary drawing.

Presentations

The /slides skill creates and updates HTML presentations. Iskra chooses the layout, fonts (Cyrillic and Chinese), color and visual techniques on its own: first it considers your preferences, references and brand style, then the audience, the source data and the key takeaway. Navigation, the slide counter, the progress bar and the buttons are styled to match. You can separately ask to change the navigation bar — for example, to gather the arrows, slide number and progress into a compact capsule at the bottom following your sample, while keeping the slide content. For data exploration, you can add period switching, option comparison, filters or expandable details; Iskra chooses interactions that are useful for the purpose of the presentation and takes keyboard, touch and screen size into account. You can describe the style you want right away or attach a sample; there is no need to pick a template from a gallery first. Iskra asks only for information without which the task cannot be done: the goal, source materials, audience, duration or number of slides.

The ready-made scenarios “Create presentation” and “Update presentation” show the stages of the work, from analyzing the task to review and publishing. A regular request in chat goes through the same process. Iskra creates the HTML in the chat working folder and checks the content, calculations and rendering itself. For a new interactive version, it reviews every slide at desktop and phone sizes, including the meaningful steps that reveal content. To do this, it uses state navigation supported by the file itself. Clicks are checked separately; Iskra states the actual scope of the check and the remaining limitations. After fixes, it rechecks the affected parts of the presentation.

When updating, Iskra works from the published presentation, keeps its design, format and navigation unless you asked to change them, and makes the edits in the same file. In the same chat, the result keeps the file name and version history. To update it in a new chat, attach the source HTML; a read-only file becomes the basis of a separate working copy. The source mode is preserved: when editing a print presentation, Iskra updates the HTML and the linked PDF. Asking to convert a presentation to another mode creates a separate version.

For a new presentation, Iskra determines the output from your request:

Interactive HTML is the main output unless you asked for print or PDF. By default, the presentation opens in the browser as a sequence of widescreen slides with buttons and keyboard controls. You can request a different format or viewing mode, such as vertical slides, scrolling or navigation through a table of contents. Animation, reveal steps and interactive elements depend on the task. In the regular show, the background and design are confined to the slide; the surrounding viewing area is visible around it, and the buttons sit compactly below it. You can specify a different design approach in your request. When it helps the task, Iskra adds data exploration: for example, choosing a period or switching between amounts and shares, with the chart and explanation updating accordingly. This mode does not promise an exact match when exported to PDF.

HTML for PDF is a separate static version: each slide has print geometry, and the data, values, labels, notes and static visuals are visible

without running any animation. Before handing it over, Iskra exports the HTML it actually created with the standard offline converter and delivers the source HTML together with the resulting PDF. This is how Iskra handles an explicit request for print or PDF.

Iskra creates both versions — interactive HTML and a verified print version — when you explicitly ask for both. An attached PDF or document used as source material does not by itself mean a print request.

The source of each version is a separate HTML file. New interactive versions work both in Iskra and after download: outside Iskra, the browser loads the same fonts, icons, charts and navigation tools from a CDN, so internet access is required. The file stays editable; you can upload it to Iskra again. New slides use Reveal.js for navigation, reveal steps and the presentation overview. Consistent Lucide outline icons are available for subject labels and buttons: their color, size and stroke width can be changed together with the slide design. Iskra first determines what information a graphic adds, then chooses its form and aligns it with the overall composition. Previously created files still open in the preview with their original fonts and navigation. A verification PDF is added to the print HTML. The skill does not create PPTX.

The HTML is created directly in the chat working folder. Iskra refines and checks the same file, and updates an existing presentation in place, keeping its name and version history. Intermediate changes are visible before the check is complete. For a read-only attachment, a working copy is created first.

In HTML for PDF mode, each slide takes up a separate 16:9 page, and all the meaning is visible at once in the static markup, without step-by-step reveals or JavaScript. If the material doesn’t fit on one slide, Iskra splits it into additional pages, preserving the meaning and a readable text size.

Example requests:

“make a presentation for a ten-minute talk based on this document”

“rework this HTML presentation for self-paced reading”

“update the figures on the third slide, keep the design and the file name”

“prepare an interactive presentation and a separate print version with a PDF”

Charts and dashboards

The /dashboard skill creates and updates interactive HTML files. When creating one, it chooses the form based on the task:

a single visualization of observations, comparisons, time series, distributions, hierarchies or flows — diagram.html;

a calendar work schedule as a Gantt chart, or an “activity on node” (AoN) dependency network — planning-diagram.html;

an analytics dashboard of several linked views, KPIs, tables and filters — dashboard.html.

For a regular chart, Iskra formulates the key takeaway, chooses a form that suits the meaning of the data and adds an exact table to the visualization. For a dashboard, it chooses the structure, typography, color, metrics and filters on its own; you don’t need to pick a theme or template in advance. In tables, numbers with units, amounts with “$/mo”, item numbers with names and column headers don’t break across lines, and a table that is too wide scrolls inside its panel. If you specified a style, Iskra follows it when creating the file and in later edits.

The skill has four ready-made scenarios: “Create dashboard”, “Update dashboard”, “Create chart” and “Update chart”. They show progress from analyzing the data or choosing the source file to checking the finished result. The chart

scenarios work both for regular charts and for Gantt/AoN — the route depends on the form you need. A regular request in chat goes through the same stages, even if you didn’t choose a scenario.

The new HTML appears right away in the chat working folder. When updating, Iskra makes the requested changes directly in the existing file and checks it again. In the same chat, the result keeps the file name and version history; earlier versions are available in the history. If several files fit, Iskra asks which one to use. To update in a new chat, attach the source HTML: if there is no file, Iskra asks for it. If the source is read-only, the result is saved as a separate working file, and the source stays unchanged.

In planning mode, an explicitly named form takes priority. Without an explicit choice, Iskra uses a Gantt chart for a plan with calendar dates, phases and milestones, and AoN for the order of work and directed dependencies without a calendar scale. If both forms fit equally well, or a Gantt chart lacks dates or durations, Iskra asks one clarifying question before creating the file. In a dated Gantt chart, it also checks the calendar endpoints of FS/SS/FF/SF links; conflicting dates or links are clarified with the user, without automatically rescheduling any work.

A planning chart shows the activities, milestones, hierarchy and FS/SS/FF/SF links with lag that you provide. Automatic calculation of CPM, the critical path, float, work calendars and holidays, resource leveling and effort calculation are not part of this output. Such values can be shown as facts from the source data with an explicit label. For AoA, a line-of-balance (cyclogram) chart or exact calendar calculations by months and years, Iskra first explains the limitation and offers a Gantt chart or AoN.

Data, styles and interactivity live in the HTML. In Iskra, charts, fonts and icons are loaded from verified platform resources inside an isolated preview. A downloaded new HTML file loads these libraries from a CDN when online; you can attach it again for viewing and editing. Links inside the HTML don’t open external sites from the preview. Remote images are blocked at first; a separate button lets you show them for the current view.

Before confirming that the result is ready, Iskra itself verifies the data and calculations, checks the source and reviews desktop and mobile screenshots, including the lower sections. It assesses readability, overlaps, text clipping, charts, fonts and icons. After fixes, it rechecks the affected parts. The screenshot check covers the main vertical scroll. Clicks, filters, keyboard and nested scroll areas are checked separately when interaction is available. If coverage is incomplete or images are unavailable, Iskra reports the limitations of the check.

Review and fixes happen in the same file, without moving it out of “Drafts” at the end. If the file changed during refinement, Iskra takes those changes into account first. If the review is unfinished or the work is interrupted, the last written version remains; you can open the previous one in the history. The file being available does not yet mean the check is complete: Iskra reports the status and any remaining issues. The skill produces interactive HTML; a separate PDF is not created automatically. If a PDF or print version is explicitly requested for a planning chart, Iskra offers to continue with the available interactive version once you confirm.

Example requests:

“show revenue and margin by month in a single chart”

“build a Gantt chart from these activities, dates and dependencies” “show an AoN network: design → development FS, development → testing SS+2 days”

“make a sales dashboard with filters by region and period, KPIs and a table”

“add a filter by manager to this dashboard, keep the design and the file name”

“update this chart: add August and keep the rest of the data”

Design and fonts for charts and dashboards

Iskra chooses the design on its own: structure, typography and color — to suit the task and the data. You don’t need to pick a theme or template in advance. If you named your own style, Iskra keeps it both when creating the file and in later updates.

Font families with Cyrillic support are available for body text, headings and numeric values, each with its own role:

Golos Text — neutral interface, analytics, body text;

Exo 2 — scientific and tech-oriented presentation;

Unbounded — a short futuristic display accent;

Jura — digital geometry and instrument panels;

JetBrains Mono — code, numbers and technical values;

Pixelify Sans — gaming and pixel aesthetics;

Old Standard TT — historical and archival presentation;

Cormorant Garamond — editorial, museum and literary presentation;

Poiret One — Art Deco and constructivist headings;

Neucha — a short handwritten accent.

When the file is opened outside Iskra, the browser loads the fonts and the charting library from a CDN — this requires internet access. Inside Iskra, the same known resources come from your installation. Previously created files still open with their original fonts.

Image generation

Creates one to four separate raster images from a text description; if you don’t specify a number, it creates one. Several variants of one job get the same description and may look similar. If you need different pictures, describe separately what each one should contain: Iskra runs a separate generation for each job. Two objects in one picture and two separate pictures are different requests. The same rule applies to different edits, videos and audio files: each piece of work has its own sources and settings. One file being ready does not cancel the other results you ordered; suitable finished files are kept while the remaining ones are being clarified. For a new composition, you can attach up to five pictures and explain the role of each: for example, the product in the first and the style in the second. To change an existing picture, you can attach a PNG/JPEG/WebP/GIF file or refer to a file in the current chat: the skill adds, removes or replaces objects, changes the background, details or style, and saves the result as a new copy. You can add up to four references to the base image; in total, up to five images and up to 10 MiB combined. The source files are not changed. If some variants fail, the finished ones remain available: Iskra shows them and explains what was not completed. Successful variants are not automatically regenerated. For a local edit, Iskra states in the instruction what exactly to change and what content outside the selected area to keep; if the edit boundary allows substantially different results, it may clarify it first. Editing is done without a mask, so pixel-perfect preservation of the other areas is not guaranteed. The result is saved to the session files and is always shown in the reply.

Example requests:

“draw a minimalist poster of a city at night in shades of blue”

“turn the attached photo into a watercolor, keep the composition”

“create a square illustration of a robot gardener for a cover”

“make three poster variants: take the product from the first picture and the style from the second”

Choosing a model. Name the model in your request to apply it only to this generation or edit. The next request that doesn’t name a model uses the saved setting. Iskra suggests options based on purpose and capabilities: languages, style, available parameters and limitations. For a permanent change, ask Iskra to save the model or open the skill settings: the “Image model” section → “Choose a model”. You save the form yourself; this setting applies to subsequent requests until you change it.

Iskra shows finished images, videos and audio with a short description. You can ask separately for details about the parameters and cost. A routine background cost reconciliation doesn’t require a separate notification; if an administrator check is needed, Iskra briefly lets you know.

Aspect ratio and resolution depend on the selected model: an unsupported parameter has to be changed before generation. Describe style and composition in words; quality, file format, transparent background, resolution and proportions can be specified explicitly if the model supports the corresponding setting. Iskra passes them as separate parameters.

The skill’s catalog marks available operations and unconfirmed capabilities. Its limits take into account both the model and Iskra’s tools: advanced features in the provider’s description do not by themselves mean they are available through the skill. The skill produces the number of variants through separate generations, not through the provider’s per-request image limit. Models without an available route are hidden from the skill’s regular answer about models; when you check a specific model, the explanation of the incompatibility is kept. Before every request, including those without references, the skill checks that the files and settings are compatible. If a model requires a reference, returns only a vector format or is intended for a different operation, Iskra explains the limitation before generation. Transferring a style to a new composition is different from editing the original picture. The selected model and your requirements are never silently replaced. New models are connected automatically within known capabilities; a new or incompletely described mode may require a skill update. In particular, new text-to-speech models still require confirmation of the audio format: the current provider catalog doesn’t provide enough information for a safe automatic choice. An “Unverified” mark on an image may mean that detailed capabilities will be clarified at request time, not that support is missing.

When an error is correctable, Iskra receives the reason and the allowed parameters so it can adjust its settings or agree with you on changing your requirement. After a request is sent, Iskra distinguishes between a content refusal, another explicit refusal by the generation service and a completed response without a finished file. Each of these outcomes may involve a charge. Iskra explains the result, keeps the files that are already finished and agrees a new paid attempt with you. If the response was lost after the request was sent, the generation may already have run and cost money. In that case, the next step is to check the known video job or discuss a new attempt with you, rather than automatically creating another result. Error explanations and continuation rules are the same for the system connection and for your own key; what differs is the access setup and the payment method.

The skill is available when the generation connection is configured for your installation. The cost is charged in MT based on the actual cost of the generation, not as a fixed amount. The private “Antonym” mode covers Iskra’s LLM requests but not the work of the generation service itself: if the skill calls an external image provider, the description and the reference are sent there even with Antonym turned on.

Video generation

Creates a short video from a text description. You can set the first frame or a “first and last frame” pair from files in the current chat. Another option is up to three object or style references, if the model supports this mode. Frames and regular

references are used in different requests; the last frame is set together with the first. The total size of input images is up to 10 MiB. The finished video appears as a link in the reply and plays in the built-in player on the artifacts panel.

Video takes noticeably longer to generate than an image and costs more — from tens of seconds to several minutes. If the result isn’t ready by the end of the current reply, the assistant says the video is being prepared and picks it up on the next turn: ask “check whether the video is ready” or just wait. The cost is determined by the paid operation that was performed; the file may still fail to be delivered.

Example requests:

“make a short video of a kitten playing with a ball of yarn”

“bring the attached photo to life: gentle movement of clouds and water”

“make a transition from the first picture to the second over five seconds”

Choosing a model. You can name a model for a single video; a video that has already started will continue with that model even if you change the settings later. For a permanent choice, use the catalog: in the skill settings, open the “Video model” section and click “Choose a model”. Duration, frame aspect ratio and reference support depend on the model; if a capability isn’t confirmed yet, Iskra offers to change the parameters or choose a verified model. The cost is charged at the selected model’s actual price at the provider; the administrator sets a cost cap per generation that protects against extreme amounts.

If the remaining limit isn’t enough to admit a generation, Iskra explains why: the balance is insufficient, part of the limit is held by pending charges, or the limit on unfinished generations has been reached. With such a refusal, the generation has not started yet and nothing is charged for the attempt. You can continue once the stated cause is resolved. If the balance is low even without pending charges, waiting alone won’t help — you need a larger limit or an administrator review of the admission threshold for this media type. This threshold is set by a skill setting: changing the model, duration, resolution or reference within the same media type doesn’t change it. This applies to all modes of the system media skill.

Text-to-speech

Generates speech from text and saves it to the session files: Gemini as WAV, other models as MP3. A link appears in the reply that you can play in the built-in player on the artifacts panel. In a single call, Gemini accepts up to 1,000 characters, and other models up to 4,000. For long text, Iskra suggests splitting it into meaningful parts; each part is a separate paid generation.

Example requests:

“read aloud: hi, this is Iskra”

“read this paragraph in a pleasant female voice”

“read the following text calmly and slowly: …”

“read this text excitedly, in the voice from the attached sample”

Give the text to be read and your delivery preference separately, in plain words. Iskra chooses how to pass the preference based on the capabilities of the selected model:

Gemini 3.1 Flash TTS, Fish S2 Pro and Fish S2.1 Pro (including Free) accept brief preferences for pace, emotion and intonation. Iskra translates only the preference into a short English phrase — up to 200 characters on a single line. The text itself stays in its original language. This is a request to the model, so the exact intonation is not guaranteed.

MAI-Voice-2 supports speed adjustment from 0.5 to 2: for example, “read it at speed 0.8”. Separate styles and emotion instructions are not yet confirmed in this integration.

If the current model doesn’t support the preference or the way to pass it isn’t confirmed yet, Iskra checks the capabilities and offers a choice: a compatible model with the desired delivery, or the current model without this setting. Saving a model in the form does not by itself cancel the intonation preference. When you agree on a choice, only the disputed setting changes; confirmed preferences, such as a slow pace, are kept. An unconfirmed setting doesn’t mean the model can’t control delivery at all. Regular speech synthesis with another model can work even without confirmed support for delivery control. The catalog shows the available settings, the format and the text length limit.

To synthesize speech in a voice from a sample, attach an MP3, WAV, FLAC, OGG or M4A file of up to 10 MiB and specify which file to use. You can add the text spoken in the sample. The sample is used for this request only and does not create a permanent voice profile. The source file stays unchanged.

Choosing a voice and model. The voice and TTS model come from the skill settings (the “Audio models” section with the same model catalog). You can name a compatible voice in your request: it replaces the saved one for this speech generation only. You can also choose a model for a single request. When you change the model for one request, the voice from the previous model is not carried over; you can explicitly choose a compatible voice. An empty voice means “Automatic” mode: Iskra chooses the first voice from the selected model’s catalog, and if there is no voice list, it leaves the choice to the provider. This is the catalog order, not a quality ranking. The automatic choice may change after a catalog update; to keep the same voice for future speech generations, select it explicitly in the settings. In the reply, Iskra states the voice actually used when its name is known. Voice from a sample has its own model setting, and the regular voice from the settings is not carried over to this mode.

If the speech doesn’t sound right — unclear, with an accent or in the wrong language — tell the assistant: it shows other text-to-speech models from the catalog and applies the chosen model to a single speech generation. To change the permanent setting, ask Iskra to save the model: a form opens that you save yourself. It applies until the next change. Text-to-speech has no separate language parameter: the model detects the language from the text, so a different language requires a multilingual model — check its description in the catalog. A voice belongs to its model: when you change the model, a compatible voice is kept. If the catalog excludes the previous voice, or the new model is in the catalog but doesn’t publish a voice list, the form switches the previous voice to “Automatic”. A catalog loading failure does not erase the voice by itself. You can choose a specific voice in the same form. If a voice name that isn’t in the catalog has already been entered manually for the same model, the form saves the setting so you can fix it, but the skill rejects it before generation. For a model without a list of named voices, choose “Automatic”; for a voice from a sample, use the separate mode. Before generating speech, Iskra checks the model and voice against the catalog. If the catalog is unavailable or the voice is incompatible, the paid generation doesn’t start: first you need to restore access to the catalog or choose a suitable voice. Fixing the voice keeps your delivery preferences. These rules are the same for the system skill and for the version with a personal OpenRouter key.

The cost is charged in MT at the provider’s actual price. If an error occurs after the request is sent, the provider may have already performed the generation. Iskra will let you know; start a new paid attempt deliberately. The text length limit reduces the risk of a file that is too large but does not guarantee its size: the duration depends on the delivery and pauses.

Music generation

Ask Iskra to create an instrumental piece or a song with vocals. Describe the genre, mood, instruments, tempo and structure: for example, intro, verse, chorus and outro. For a song, you can provide your own lyrics; the vocal language is set by the lyrics and

your preferences. You can attach one image to the music description as a mood reference. An instrumental piece is created without lyrics. The description and lyrics together can be up to 4,000 characters.

Example requests:

“Create a calm instrumental piece: piano, soft strings, evening mood, smooth ending”.

“Make a song in Russian from these lyrics, acoustic guitar and soft vocals; mark the verse and chorus: …”.

“Use Lyria 3 Clip just this once; keep the saved model as it is”.

Lyria 3 Clip is designed for 30-second clips. Lyria 3 Pro is suited for songs with structure; the duration is passed as a preference. If you need a strictly defined running time, state it separately: the capabilities of the selected model have to be agreed before generation. Creative preferences for the performance are judged by the finished result.

The music appears in the chat’s regular audio player, is saved to files and is available for download. One call creates one mixed-down audio file. The skill requests MP3, but the provider may return WAV or FLAC; the actual format is shown by the file extension. If you need a strictly defined format for another application, a separate conversion may be required. This skill does not output separate instrument tracks. You can name a model for a single generation; for a permanent change, ask Iskra to save the choice or open “Music model” in the skill settings. These features are also available in the version with your own OpenRouter key. Through the form Iskra offers, you can save a model with confirmed music support. If an error occurs after the request is sent, the provider may have already done the work: first check the files in the current chat, and treat a repeat generation as a separate decision. This rule also applies to an explicit refusal by the service or a completed response without music. Unlike video, this skill has no check for a pending music job.

Audio and video transcription

Attach a recording to your message and ask for a transcription. Iskra uses the shared audio and video transcription tool — where the administrator has enabled it. A long recording is processed asynchronously, and the full recognized text is saved to the chat’s “Drafts”. If the tool is unavailable in the current chat, Iskra will tell you.

Example requests:

“transcribe this audio”

“convert the voice message to text, the language is Russian”

The media skill no longer has a separate speech recognition setting. The 10 MiB limit for a voice sample does not impose any limits on the shared transcription tool.

A request to transcribe a recording with timestamps and speakers is handled by the platform transcription service — see “Audio and video transcription”.

Reputation.ru case law search

This system skill searches commercial (arbitrazh) court cases, cases of courts of general jurisdiction and decisions of the Supreme Court of the Russian Federation, retrieves the texts of the documents it finds and runs an in-depth analysis of case law. The administrator sets the corporate API key, and the cost of external requests is converted into MT. The user’s request is sent to Reputation.ru. Capabilities, pricing and the personal key option are described in a separate section.

DataNewton

A skill in the “Integrations” category that checks Russian counterparties through the DataNewton service using your API key: it finds a company or sole proprietor by name, INN (taxpayer ID) or OGRN,

and shows the company profile from the EGRUL/EGRIP state registers, financial statements, risks and scoring, commercial court cases, government contracts, affiliations, signs of bankruptcy and inspections. Read-only — the skill doesn’t change any data in DataNewton. How to get a key and what the skill can do are described in a separate section.

Yandex Metrica

A skill in the “Data” category that answers questions about your website traffic using your personal Yandex OAuth token: counters, visits and visitors for a period, traffic sources, goals and conversions. Read-only — the skill doesn’t change Metrica settings. How to get a token and what the skill can do are described in a separate section.

Yandex Webmaster

A skill in the “Data” category that answers questions about your website in Yandex search using a personal OAuth token: verified sites, SQI (site quality index), pages in search and excluded pages, Webmaster issues, popular search queries with positions, and trends in impressions, clicks and SQI. Read-only — the skill doesn’t change sites or their settings. How to get a token and what the skill can do are described in a separate section.

Wildberries Seller

The connection is read-only: it retrieves product cards, current stock in WB warehouses and seller warehouses, FBS assembly orders and their current statuses, prices and discounts, as well as preliminary operational sales and returns for a recent period. It doesn’t change any data in the Wildberries seller account.

Find Wildberries Seller in the catalog, install it and open the skill. In the “Wildberries — read-only connection” form, fill in your personal connection:

1. In WB Partners, open Profile → API integrations → Create token.

2. Choose a personal or service token, the categories Content, Analytics,

Marketplace, Prices and discounts, Statistics, and the “Read only” access level.

3. Paste the value into the “Wildberries token” field. If it is a Service token, also fill in

“X-Client-Secret — Service token only”; for a personal token, leave the second field empty.

Both values are stored only in your Iskra profile. Even if the skill is installed for an organization, each member connects their own seller account separately; there is no shared token for the organization or the whole platform. Don’t paste tokens into the chat. After updating from an old version, enter the token in the new form again: the former shared slots and old profile slots are safely deleted and not carried over.

Operational sales and returns are preliminary, available for the last 90 days at most, and do not replace accounting reconciliation. The size of the response and of each page is limited; if there is more data, ask Iskra to continue from the next page.

First, find the product card by name, vendor code or barcode: the result gives you the nmID and chrtID for stock queries. When looking for a specific size in WB warehouses, Iskra passes the chrtID together with the corresponding nmID. For seller stock, first get the list of warehouses and choose the warehouseId. For statuses, first get the list of FBS assembly orders, then request statuses for the returned IDs.

If the connection doesn’t work:

Token not configured — open the secure form and add it.

401 — the token is invalid, revoked or expired; replace it.

402 — the seller account denied access to your account; check the account’s access and the required API category.

403 — add the required token category or check permissions; for a Service token, check the X-Client-Secret entered in the same form.

413 — narrow the filter, page or period.

429 — wait and retry the request.

Temporary Wildberries outage — retry the request later.

Ozon Seller

The skill connects an Ozon seller account and only reads data: products, stock under the FBS, rFBS and FBP schemes, prices, FBS or FBO orders and basic sales for the selected period. FBO stock is not included in list_stocks. A single request for orders or sales covers no more than 31 days; for a longer period, ask Iskra to process several windows one after another. Order pages continue via the cursor from the previous response. Fresh analytics data may appear in Ozon with a delay, and Ozon doesn’t send days without sales at all — so the response lists only the days that have data, and the boundaries of the requested period are shown in separate fields.

If the skill catalog is enabled on the platform, find Ozon Seller in the “Integrations” category and install it for your personal profile or organization. Without the showcase, the skill remains available for an explicit call via /. In Ozon Seller, open Settings → API keys and create a key with access to the data you need. In the “Ozon — read-only connection” form, fill in two different values: “Ozon store Client ID” — the numeric ID next to the key — and “Ozon Seller API key” — the secret value of the key itself.

The connection is stored only in your Iskra profile. Installing the skill for an organization doesn’t create a shared key: each member enters their own credentials. Don’t send them in the chat. The connection check confirms access only to the product catalog; permissions for prices, stock, orders and analytics are checked when you make the corresponding request.

If Ozon was previously connected with a shared organization key, each member needs to create a personal connection after the update. An already configured personal Ozon profile is kept.

If the connection doesn’t work:

Key not configured — open the secure form and fill in both fields.

ozon_auth_failed — check the Client ID and recreate the API key.

ozon_scope_denied — grant the key permission for the specified operation.

ozon_rate_limited — wait and retry the request.

ozon_unavailable — try again later: the Ozon API is temporarily unavailable.

ozon_timeout — Ozon didn’t respond in time: reduce the page size or narrow the period; retrying the same request usually doesn’t help.

ozon_response_too_large — the response exceeds the safe size: reduce the page size or narrow the filter.

ozon_egress_blocked — requests to Ozon are blocked by the platform settings; contact your administrator, retrying the request is pointless.

ozon_endpoint_missing — Ozon no longer serves the required API address; tell your administrator, the connector needs an update.

bad_input or ozon_bad_request — check the period, the FBS/FBO scheme, the cursor and the filters.

ozon_provider_error — read Ozon’s safe message and fix the parameters it mentions.

ozon_response_invalid — try again later; if the error persists, tell your administrator that Ozon’s response may have changed.

settings_backend_error — try again; if the error persists, contact the platform administrator.

If a built-in skill is unavailable or its mode can’t be changed, the platform administrator has changed the policy.

The platform also ships internal system skill sets, for example for feedback, checking answers against the help and creating text skills from chat. They may not appear as regular user skills, or they may need to be enabled by the platform administrator.

Overview

“Task Advisor” is a system skill that sometimes adds a short follow-up at the very end of Iskra’s reply: one or two items under the heading “I can also:”. It suggests the next useful step in your task — for example, checking the result on real data or turning the reply into a table.

There is nothing to configure: the skill has no keys or parameters. On the Skills page (/skills), it appears in the “System” section as a separate row with a single toggle. It doesn’t have the usual “Off / On command / Auto” choice — it isn’t a tool and can’t be called with the / command.

When the suggestion appears

The platform administrator sets the length of the automatic period — either by the number of your first messages in the account or by the number of days since sign-up. Until the administrator changes this setting, the default applies — your first 20 messages in the account. The count is shared across the account: it doesn’t start over in a new chat and isn’t reset when you switch between your personal profile and an organization.

Only regular messages in personal chats count — both in your personal profile and in an organization. Saved legacy group-chat sessions in projects (new messages can no longer be sent in them), skill runs via the “Run” button, scenario runs and API calls made without the interface don’t count toward this total and don’t get the suggestion.

After the automatic period, the suggestion is no longer added on its own. If you still want it, turn it on with the toggle.

Even within the automatic period, the block doesn’t appear in every reply. It won’t appear if Iskra first needs to ask a clarifying question, if the reply ended with an error or was interrupted, if the reply is a refusal for safety reasons, or if you are saying goodbye or asking it not to suggest a follow-up. If you gently corrected the result, Iskra first fixes it and adds at most one short item — about checking the fix or the next immediate step. With strong or repeated dissatisfaction, there is no block at all, and it returns once the conversation continues normally: irritation is not remembered as a user trait.

A created file, table or document doesn’t cancel the suggestion — the block remains the regular end of the same reply.

The toggle

The setting applies to the whole account — across all your profiles, personal and organization. Next to the toggle, one of three labels is shown:

“Turns on automatically” — the initial state. It doesn’t mean “always on”: this is the name for the automatic period, whose length is set by the platform administrator (by default, the first 20 messages in the account).

“Turned on by you” — you turned the suggestion on yourself; it is added even after the automatic period.

“Turned off by you” — you turned the suggestion off, and it isn’t added.

The toggle records only “on” or “off”. You can’t return the card to its initial automatic state from the interface: after the first click, your choice applies.

Explanations sometimes appear under the toggle:

“Applies automatically to your first 20 messages in this account” — you are within the automatic period. This is the label for the default value; if the platform administrator has configured the period differently, the number will be different, and in “by days” mode the label refers to days since account sign-up instead of messages.

“The automatic mode for this account is not fixed yet” — the automatic behavior hasn’t been determined for your account yet. This is a neutral note: it doesn’t mean that anything will turn on later.

“The skill is temporarily switched off by an administrator” — the platform administrator has paused the skill. This isn’t a switch-off on your side: your choice is kept, but the toggle is unavailable for now, and the suggestion isn’t added for anyone.

“There is nothing to apply right now” — a rare system situation: there is no suitable version of the suggestion, so the toggle is locked.

Overview

For recurring work, start with a text skill: an instruction that Iskra follows using the available tools and materials. Such a skill suits analysis, calculations, documents and multi-step processes. If there is no connection to the service you need yet, describe to the developers the service itself and the actions you need: what data to retrieve and what to write.

The fastest way to create your own skill is to ask the assistant right in the chat. If the conversation has produced a useful instruction, response template or procedure that you want to reuse, say, for example:

“make a skill that …”

“remember this as a skill”

Iskra uses what it already knows from the conversation. If the task is still unclear, it asks one or two short questions: what the result should be, where to get the data, and which materials or services are needed. You can show one example of a typical work task. If the instruction is already complete, Iskra immediately offers to save it.

What to tell Iskra when creating a skill

Start with a plain description of your work. Iskra will help you clarify whatever is still missing:

Task and recipient: “What do you want to do each time? Who will read the result — you, your colleagues or a client?”

Finished result: “Do you need an answer in chat, a Word document or an Excel spreadsheet?”

Materials: “Where are the current prices and terms? Is there a sample of a good proposal?”

Connections: “Where should the deal details come from? Which mailbox should the approved email be sent from?”

Review: “What needs to be checked before delivery? Who approves sending?”

Who can use the skill: “Do you need it for yourself or for your team?”

These are examples of topics to discuss. Iskra takes information it already knows from the conversation and asks one or two questions per message. If the information can be taken from a document, just show the file: Iskra will figure out the column names and the date format in a spreadsheet on its own. It will also suggest a name for the skill and a description of when to use it.

For example, you can write right away:

Create a personal skill for commercial proposals. Take client data from amoCRM. The Excel price list, delivery terms and Word template are in the “Sales” Memory folder. The output should be a Word document. Before delivery, a separate helper should check the prices and terms. Send the email via Yandex Mail after I confirm the recipient, text and attachment.

Structure of a text skill instruction

A useful working procedure is easy to describe in six parts:

1. When to use: the task and the signs that the skill fits.

2. What to clarify: missing data, the recipient and conditions; anything known from the conversation or

from available documents is used by Iskra right away.

3. Sources: connected services and materials, the purpose of each file, and what to do

when a source is unavailable or outdated.

4. Workflow: the steps of analysis, calculations and preparing the result.

5. Review: what to check against the source data, when to bring in a separate helper and

how to address its comments.

6. Result: the format, where to save it, and the actions the user

confirms before sending or writing to an external service.

For example: “Every month, compare the website metrics with the previous month. Clarify the site and the period, get the data from Metrica, check the calculations, prepare a report with charts and save it to Chat files”. For regular execution, schedule a run in Plans; the text skill describes the work itself.

Where to keep skill materials

The wizard in chat saves the instruction text. A short sample and a task for the reviewer can be included in this text; separate shared files are provided in an uploaded ready-made bundle. Company materials are connected from Memory.

What to keep Where to keep it Example

Steps for preparing a report,

Workflow Skill instruction review rules and the format

of the result

Anonymized sample

Shared files of a ready-made Bundle of an uploaded

report, task for

skill skill

the reviewer

Current price list,

Company materials Memory folder attached

delivery terms,

needed across chats to the skill

branded Word template

Message attachments

Documents for a single client or a single task or a separate folder connected to the chat Client contract, order contents, meeting notes

via “@”

What to keep Where to keep it Example

First draft

Intermediate versions Drafts of the proposal, calculations and

the helper’s comments

Final result and important notes Chat working folder, “Chat files” section Reviewed proposal to send to the client

Connection keys and passwords Secure connection settings App password for email, service token

You attach a Memory folder on the skill card. Put the materials you need in a separate folder: when using the skill, Iskra can read all of its contents. Attaching gives read access; to prepare the result, Iskra works with copies of the documents. For an organization skill, the folder is attached by the owner or an administrator, and colleagues need access to its materials.

It helps to explain the purpose of each material: “Take prices from the price list, deadlines and discounts from the delivery terms, and formatting from the template”. If a document you need is unavailable, Iskra can ask where to get it.

Drafts are suitable for temporary work. They are kept for 30 days after the last change to their content; viewing and renaming don’t extend this period. Ask Iskra to move the final document and any comments you want to keep to “Chat files”.

A skill bundle combines its instruction and additional files. It is the right place for shared materials for all users: a sample report with fictional data, a general proposal outline or a separate task for the reviewer. A specific company’s price list and branded materials are better kept in Memory, where they can be updated separately from the skill.

A bundle with separate files is uploaded as a ready-made skill in the Skills section. Saving an instruction in the wizard doesn’t attach a Memory folder and doesn’t add such files to the bundle.

How to set up a review of the result

For an important document, you can ask for a review by a separate Iskra helper. Give it a self-contained task with clear criteria. For example:

Compare the commercial proposal with the client’s request and its clarifications. Check the order contents, prices, discounts and deadlines against the price list and delivery terms. Open the document and assess whether the tables are easy to read. For each comment, indicate the place in the document and the source it conflicts with. Separately list any information that is missing for the review.

Iskra includes the workflow in the skill instruction: prepare a draft, give the helper the original request, clarifications, materials, the document and the review task, then address the comments and recheck the changed result. For a short answer, you can choose a simpler check, for example a self-review of the text before sending.

The helper can also make mistakes or miss a discrepancy. Ask it to state which information it was able to check and what was missing. A quality check doesn’t replace your approval of external sending: the recipient, the email and the attachments are confirmed separately.

After creating the skill, try it on two or three familiar work tasks. Check whether the right materials were used and whether the result is convenient to work with. If something

doesn’t fit, ask Iskra to change a specific step of the instruction. For a targeted edit, it uses the current text and what it already knows.

Sample skills and the materials they need

The folder and file names below are only examples. The person using the skill must have the service connections available and set up.

Sales proposal — amoCRM and Yandex Mail

Task: prepare a proposal for the “North Office Fit-Out” deal for 20 workstations and send it once it is approved.

Connections: amoCRM for the deal and contacts; Yandex Mail through the “Mail” skill to send the email.

Materials: the “Sales” Memory folder with an Excel price list, delivery terms and a Word proposal template; the order contents are in the user’s message.

Result and review: the calculation and the first version go to Drafts; a separate helper checks the prices, order contents and terms. The finished Word document goes to “Chat files”. Before sending, the user confirms the recipient, the email and the attachment.

Website report — Yandex Metrica

Task: compare two months for a selected website and show how traffic and the share of leads changed.

Connection: Yandex Metrica; the request must name the website, the months and the year.

Materials: the “Website analytics” Memory folder with a definition of what counts as a lead, the ad campaign calendar and a previous report to use as a model.

Result and review: calculations in Drafts, the finished report with charts in “Chat files”. The helper checks the periods, the figures and the grounds for the conclusions.

Meeting prep — Bitrix24

Task: prepare for a meeting with the company “Vector” about renewing the service: recall what was agreed and collect the open questions.

Connection: Bitrix24 for the deal, contacts, comments and scheduled activities.

Standing materials: the “Renewals” Memory folder with pricing plans, rules for approving terms and a list of required questions for the meeting.

Materials for this chat: the current contract with “Vector” and notes from the previous meeting — attached files or a folder connected via “@”.

Result and review: a short brief with the agreements, open questions and the next step. A separate helper checks the agreements against the correspondence and notes and separates proposals from confirmed facts. Iskra addresses the comments; the user reviews the brief before the meeting.

Reply to a customer in the company’s style — Gmail

Task: prepare a reply to an email about a delivery delay and suggest the next step in the company’s house style.

Connection: Gmail through the “Mail” skill — the selected email and the relevant thread.

Materials: the “Customer communications” Memory folder with communication guidelines, product names and anonymized examples of approved replies; the confirmed status of the specific order is in this chat’s materials. The skill bundle keeps the general workflow and the reviewer’s brief.

Result and review: versions and comments go to Drafts. The helper checks the reply against the customer’s question, the company guidelines and current data. If the new delivery date is unknown, the reply says so plainly. After corrections, the final text

is shown to the user; before sending, they confirm the sender mailbox, the recipient and the email content.

Saving the skill

Once the workflow is clear, Iskra shows a ready form with the fields already filled in:

Name — what the skill will be called.

When to use — the description Iskra uses to decide that the skill fits a request.

Instruction — the text the assistant will follow.

Where to add — a personal skill or an organization skill (the “Organization” option is available to the organization’s owner or administrator). The assistant preselects the level based on your request, and if the request doesn’t make it clear, asks you first.

Check the values and adjust them if needed, then click “Create” — or cancel if you change your mind. The form always shows which profile the skill is being created in, so there are no surprises. After creation, the assistant states the actual publication level: if you asked for the organization level but don’t have enough permissions, the skill is created as personal — the assistant says so directly and explains how to transfer the skill to the organization.

Attached files

If you attached a file to the chat (a checklist, a procedure, a reference guide) and ask to turn it into a skill, Iskra shows what happens to each file and never silently ignores one. All active files you attached in the current chat count, including those from earlier messages.

A text file can be included in full, used selectively, or left out of the instruction text with a stated reason. For example, a price list that gets updated can stay in Memory: Iskra explains where to put it and how to connect it to the skill. With selective use, the form warns that the completeness of the extract is not checked automatically and asks you to confirm the decision separately. The same confirmation is required for a file left out of the instruction text. Full inclusion never truncates the content: if the resulting text doesn’t fit in the skill, creation stops and Iskra offers to shorten the instruction or pick the useful part.

In a program skill, an attached .go file can be saved byte for byte as source code if it really is part of the program build, and a small text data file can be saved as a connected resource. Binary and media files can’t be carried over directly this way: Iskra shows the limitation and asks you to confirm either selective use of the already extracted content or excluding the file.

Once you confirm, the version is sent for moderation. Iskra reports its number and current state, and you can follow the status in the Skills section. Until it is approved, a new skill is inactive; after its first successful publication it gets the “Auto” mode. If the platform administrator has turned moderation off, creating and updating text skills is blocked — an unreviewed version is never published.

Personal skill or organization skill

In a personal profile, a skill is always created as personal — only you can see it.

In an organization, the owner or an administrator can choose where to create the skill: personal (just for themselves) or an organization skill (available to all members). A regular member always gets a personal skill, even if they ask for a skill for the whole organization — the assistant says plainly that the skill was created as personal and suggests transferring it to the organization.

Text and program skills from chat

For a work process — preparing documents, doing calculations and working through existing connections — Iskra creates a text skill. Company materials can be kept in Memory and used across chats.

A program skill is needed to connect to another service: to fetch or send data. Such connections are usually added by a development team. If you explicitly ask it to develop an integration, Iskra can prepare and submit its code from the chat. Both kinds of skills go through moderation; a program skill is also built. If the connection you need doesn’t exist yet, Iskra explains what has to be added; for a suitable task, it suggests starting with a data export that you provide.

You can also update your skill from chat: ask the assistant to change an existing skill — it takes the current content, makes the edits and sends a new version for moderation. Until that submission is complete, you can’t submit another update of the same skill; Iskra suggests checking the status in Skills.

If a skill needs access keys (secrets) for an external service, the assistant first offers to open a secure form right in the chat (see “Configuring skill parameters and secrets from chat” below) or points you to the Skills section ( /skills ) — this way the secret never enters the conversation. You can also send the value as plain text, but then it stays in the chat history — see the same section for details.

More flexible methods (a wizard with parameters, uploading a ready-made skill) are still available in the Skills section — see “Submitting your own skill” below.

The number of skills you can create this way is limited so that the catalog doesn’t get cluttered. If you reach the limit, delete the skills you no longer need on the “Skills” page.

MCP

MCP is the Model Context Protocol: a common way for services to describe their tools to assistants. MCP support in Iskra is available on platforms where the skill system is enabled. When the skill system is turned off, MCP connections through skills are unavailable. Iskra works with MCP only through an available, configured program skill that describes the tools and the connection to the external MCP server in advance.

You don’t need to write such a skill by hand: send the server link right in the chat and ask Iskra to connect it. Iskra queries the server itself, sees its tools and builds a skill around them. The address, protocol and key stay on the platform side: they don’t enter the conversation, and you won’t have to write any connection code.

A public server takes one turn. Give the link and say you want to connect this MCP server. Iskra queries it and shows what it can do: the server name and version and the list of tools with descriptions. Then it proposes a skill with the tools you need — review the selection, confirm, and after moderation the server’s tools work in chat like a regular skill.

Checking an external API. To find out whether a public API responds over HTTPS, send the address and ask Iskra to check its HTTP status. Iskra can show the actual status and the part of the response available without an access key. For example, a 401 or 422 response means the server answered, but the API request itself needs authorization or different parameters.

A server with a key takes two steps. If the server requires authorization, it says so when queried — “key required” — and the tool list stays closed. In that case, Iskra first submits a skill that can only fetch the tool list and asks you to enter the key in a secure form: the value goes straight to storage, bypassing the conversation.

Once the key is entered, Iskra requests the list on your behalf and updates the skill, adding the working tools to it. The second step is mandatory: until it is done, the skill can only show the list. If Iskra didn’t offer it, ask it to update the skill yourself.

The skill subscribes to a specific set of the server’s tools. If the server can do more, Iskra can show that, but it will only call what is listed in the skill: to add the rest, ask it to update the skill. Iskra reads third-party tool descriptions and server responses as information about an external service, not as instructions to itself. Like any skill created from chat, the connection goes through moderation — until it is approved, the new skill is inactive.

Check which skills and controls are actually available in the Skills section ( /skills ). Then use only the action shown there: it defines the available way to add or configure things in your particular installation. If there is no suitable skill or action there, contact the platform administrator.

The address (URL) of an external MCP server is one of the parameters of such a connection. On its own, the address is only a connection parameter, not a ready skill: to work, you need a program skill that describes the tools and the connection. That is exactly the kind of skill Iskra builds from the link — which is why the connection is always visible in the Skills section, with a name, a set of tools and, if needed, its own key field.

A connection to an external MCP server always goes over the network. A typical local MCP server installation — a separate process using stdio (the default setup for Playwright MCP, for example) — can’t be reached from a user skill: a skill can’t start a local process. Such a server needs a ready-made supported skill or a separate network connection allowed by the platform administrator; Iskra itself does support MCP.

Whether a particular skill is available depends on its installation, settings, secrets and the permissions of the current profile or organization.

Configuring skill parameters and secrets from chat

You can ask right in the chat what settings a skill has — for example, “what are the settings of the Weather skill” or “show the 1C connection”. Iskra shows the regular parameters (server address, login, checkboxes and the like) with their current values, and for keys and passwords only whether they are set: Iskra never shows the secret values themselves, not even to you. If the skill has no configurable parameters or secrets, Iskra says so.

You can also change a regular parameter right in the chat: say what to change — for example, “change the server address of the 1C connection to …”. Before saving, Iskra shows exactly what will change and asks you to confirm with the “Save” button. If the group is configured at the organization level, only the owner or an administrator can change it from chat — just as in the skill settings window (see “Settings groups” below).

Secrets (keys, passwords, tokens) can be set the same way — confirming with a button — or through a secure form. If a skill needs access keys for an external service, the assistant can offer a “Configure skill secrets” button right away, with the skill name and a lock icon, below its reply. The button stays visible when the “Timeline” is collapsed, and clicking it opens the secure form over the chat. Parameters you discussed, such as a subdomain, may already be filled in. If you closed the window, reopen it with the same button. Values entered in the form go directly to secure storage: the model can’t see them and they aren’t saved in the conversation. This is the safest way to hand over a secret — especially for critical keys.

You can also send a secret in a regular text message on your own initiative. In that case, Iskra also asks you to confirm saving with the “Save” button and, after saving, warns that the value has already remained in the chat history and in the model context.

For critical keys, the secure form above is still recommended — with it, the value never enters the conversation at all.

If the skill doesn’t need secrets, the form tells you so.

The regular screens give the same result. In the Skills section ( /skills ), a button on the skill card opens the “Skill settings” window over the list. On the skill’s own page (the “Details” button on the card), the same fields are on the “Settings” tab.

For complex integrations, before building the skill the assistant first walks you through several clarifying steps of a disciplined engineering process.

If a key in the skill settings window is marked as “declared incorrectly in the manifest”, that doesn’t mean an administrator has to set it up: the error is in the skill’s manifest itself. Ask Iskra in chat to update the skill — the new version will fix the secret declaration.

Overview

This section applies only if skill scenarios are enabled on your platform. If the feature isn’t available, skills still work as usual — just without saved scenarios; the slash menu and the “Scenarios” tab don’t show them.

A scenario is a saved sequence of steps on top of a regular skill: for a recurring task (for example, “warehouse stock levels” in an integration skill), Iskra doesn’t have to figure out from scratch each time how to do it — it follows steps that have already been verified.

How to create a scenario. In chat, describe to Iskra a task you do regularly and ask it to turn the task into a scenario — for example: “let’s create a scenario: [DESCRIBE the task]”. Iskra clarifies the details and runs through the task with you in practice (a rehearsal), and after a successful run saves the steps under a clear name and description. If the rehearsal fails, the scenario isn’t saved — Iskra tells you so honestly.

How to run a saved scenario. There are three equivalent ways:

with a phrase in chat — for example, “run the scenario [ENTER the name]”;

with a scenario chip from the slash menu: start typing with “/” and pick the scenario you need in the “Scenarios” section — the “Scenario: <name>” chip is pinned below the input field and applies to the next message (see “Choosing a skill or scenario” in the chat article); with the “Open chat with scenario” button on the “Scenarios” tab in the skill details ( /skills → “Details” → “Scenarios”) — it opens a chat with the scenario already pinned.

While a scenario is running, a live checklist of its steps appears above the input field — see the description of how the agent works for details.

Managing your scenarios. The same “Scenarios” tab shows all of the skill’s scenarios available to you: your own, the ready-made scenario from the skill author and — if you work in an organization — scenarios published by your colleagues. The skill author’s scenario is available to everyone who has the skill connected, regardless of organization. Your own scenarios have a “personal” or “organization” label next to the name. You can rename and delete your own scenarios; in addition, a personal scenario can be published to the organization with the “Publish” button, and a published one can be taken back with the “Unpublish” button. Other people’s scenarios can only be run.

Overview

Much of what the “Skills” section offers can be done without opening the page — just ask Iskra in the chat:

“show my skills” — a list of your skills with their mode and level, to find the one you need;

“turn on the … skill on command” / “turn off the … skill” — changes the mode;

“update the … skill” — submits a new version of a text instruction skill for moderation, straight from the chat: Iskra first reads the current text of the skill and makes only the requested edit, so changes made in other chats aren’t lost (for code skills, see “Text and program skills from chat” above);

“delete the … skill” — deletes a personal skill (or an organization skill if you are the owner or an administrator), with confirmation;

“share the … skill with colleagues” — transfers your personal skill to the organization, with confirmation (see “Transferring a personal skill to the organization” below);

“what skill submissions do I have” — the status of build and moderation submissions, and cancelling a submission.

You can also find and install a ready-made skill from the marketplace, submit your own skill for publication, withdraw a submission or remove a skill from the marketplace right in the chat — see Skill Catalog for details.

Ready-made phrases for all of these actions are collected in the tip that appears when you open the “Skills” section (see above): clicking a phrase opens a chat with that text already filled in.

The “skill installed / configured” card

When a skill is installed or configured right in the chat (installation, parameter changes, saving secrets and similar actions from the list above), Iskra shows a confirmation card in its reply: the skill icon and name and an “installed” or “settings saved” mark. If the skill has ready-made scenarios, the card offers quick-launch buttons right away — they open the scenario you need without searching through “/ Skills”; if the skill has secrets, there is also a “Configure” button (with a lock icon) that opens the secure form.

No more than one such card per skill is shown per turn: for example, if you install a skill and then immediately configure it, the card reflects only the final state rather than two nearly identical entries in a row.

When you no longer need the card, close it with the “Close” cross to the right of the name — it is deleted permanently, including from the chat history and its export. Closing can’t be undone, but the action itself (installing or configuring the skill) is not undone — only the notification about it disappears.

Overview

Besides the built-in skills, you can add your own. Click the “Add your own skill” button above the skill list — a window opens where you choose the method. There are four methods, from the simplest to the most flexible:

Text skill (no code)

The simplest method. A text skill is an instruction written in plain text: you describe what the assistant should do and when, and it follows that instruction. No programming is needed.

Set a short name and description for the skill, then add one or more “tools”: for each one, specify when to use it and what to do. This is handy when you need to teach the assistant your procedure, a reply template or a sequence of steps.

Describe the capabilities the assistant needs in plain words (“generate an image from a description”), not with internal tool names of other skills such as skill__… __… . Such a reference won’t work, and Iskra won’t accept the submission — this applies to all of the methods below: the form, import and archive upload. The form shows the hint “Remove the internal reference to another skill’s tool and describe the capability you need in plain language”; after fixing it, submit the skill again.

Importing a ready-made skill (from GitHub, a folder or an archive)

If you already have a ready-made skill in the Agent Skills format (a SKILL.md file with instructions) — for example, from a public GitHub repository — you don’t need to retype it by hand. On the submission start page, choose the “Import a ready-made skill” card and specify the source:

a link to a public GitHub repository — in the form https://github.com/owner/repository , optionally with a branch and a subfolder ( .../tree/branch/folder ). A direct link to a file ( .../blob/.../SKILL.md ) is not supported;

a skill folder from your computer;

a ZIP archive with the skill contents.

Iskra finds SKILL.md itself (or AGENTS.md / CLAUDE.md if there is none), builds a draft text skill and shows it to you for review together with a report:

Source — for GitHub, Iskra shows the repository address and the commit pinned by the server. For an uploaded archive, it shows the SHA-256 computed by the server; the file name and the “folder/archive” choice are treated only as information supplied by the user and do not prove where the file came from.

Review notes — for example, detected secret patterns (such strings are automatically replaced with [REDACTED:secret] ) or mentions of external addresses.

Not imported — files Iskra discarded: scripts and executables, files with secrets ( .env , keys), and auxiliary files that a text skill doesn’t use.

You can freely edit the draft — the name, description, version and instruction text — and submit it. From there, the submission goes through the usual moderation path like any text skill; the moderator also sees where the skill was imported from.

The import security rules are fixed: code from the source is never executed, secrets are never imported, private repositories are not supported, and the source size is limited to 10 MB. If the repository contains several SKILL.md files, Iskra asks you to choose the one you need.

Program skill with AI help (wizard)

A program skill is needed for a new connection to a service: to fetch data from it or send it a prepared result. Such a connection is usually added by a development team. If you decide to build it yourself with an AI assistant, the built-in wizard helps you put together the brief: specify the service and the actions you need. Hand the resulting brief to the AI assistant, then upload the finished skill using the “Upload ready” method. Describe the workflow for using the connection — for example, preparing a report and getting approval before sending it — in a text skill.

Uploading a ready-made skill

This path is for those who already have a finished skill bundle: a program skill, or a text skill with shared resources. Attach the manifest.yaml or manifest.yml descriptor file and the source.zip source archive, then send them for review. This form accepts only a source archive; you can’t submit an OCI image through it. If you need a container skill, contact the platform administrator — such uploads are handled separately. If you only have an instruction without a finished bundle, choose the “Text skill” method.

You don’t need to upload an icon separately. To have an image on the card, put an optional icon.png in the root of source.zip — the name and letter case matter. A static PNG of exactly 128×128 pixels and no larger than 32 KiB is supported; SVG, WebP, JPEG and animation are not. Without an icon, you can submit the skill as usual. If the file doesn’t meet the requirements, the whole submission is rejected: fix it and submit again. The error message reads: “icon.png must be a static 128×128 PNG no larger than 32 KiB.”

Draw the icon in a single color on a transparent background — that way it becomes part of Iskra’s design: the platform recolors the drawing for the light and dark themes and places it on a common backdrop. This is how the built-in skill icons are made. A colored icon is accepted too, but it is shown as is — as its own image on the card — and may look like a bright spot in the dark theme. An icon counts as single-color when the red, green and blue components are equal for every visible pixel and the background is fully transparent: gray and semi-transparent edges are fine, any color tint is not.

What happens after submission

Every submitted skill goes through moderation before it becomes available:

1. The submission is sent and goes to automated review.

2. If everything is in order, the skill is approved and built.

3. After a successful build, the skill is published and appears in the catalog.

After its first publication, a personal skill gets the “Auto” mode. On the “Skills” page, you can switch it to “On command” or “Off”.

If you need a skill for a specific task but would rather not build it yourself, the submission window lists the Iskra team’s contacts: you can ask them to develop the skill for you. Ready-made skills from other users and from Iskra itself can be installed from the Skill Catalog if the platform administrator has enabled it — see Skill Catalog.

Submission statuses

Right after you submit, Iskra automatically shows the “My submissions” screen — it shows what you sent and what stage each submission is at. Later, you can open the full status from the “Skills” section: for a submission that isn’t published yet, click “Details” on the card; for an update to a published skill, click the “Open my submissions” icon on the card or the link next to the status on the skill page. The list refreshes automatically while there are submissions in progress. The possible states in plain words:

Submitted — the submission has been received and is waiting for review.

AI check — the skill is being checked automatically.

Awaiting manual review — the automated check has passed the submission to a moderator (for example, it isn’t sure of the result, or the check is temporarily unavailable). You just need to wait.

Approved — the skill has passed review.

Build — the skill is being built before publication.

Published — the skill is ready, has appeared in the catalog and, on its first publication, got the “Auto” mode.

Needs changes (rejected) — the skill did not pass review. The reason is shown below the status; after fixing the problem, you can submit the same version again.

Build failed — the skill failed to build. Increase the version number and upload it again.

Cancelled — you cancelled the submission yourself; the version is freed up and can be submitted again.

The main catalog groups these stages into short statuses. The full chain, the reason for a rejection or error, and the cancel action are always available in the “My submissions” section.

While a submission is at an early stage (submitted, under check, awaiting manual review), you can cancel it with the “Cancel submission” button on the card. After cancellation, the version is freed up for resubmission.

Occasionally the “Needs changes (rejected)” status appears not because of the skill itself but because publication failed several times in a row for technical reasons on our side. In that case, re-uploading the same file with the same version changes nothing — the submission stays rejected. To try again, increase the version number or change something in the skill file (for example, the description) and upload it again — the submission will then go through review from the start.

Skill details

The “Details” icon button opens the skill details page: the description, the list of tools, access keys (secrets) and — if the skill is yours — its content (code or instruction text). The actions available to you are gathered at the bottom of the “Overview” tab under “Skill management”. If a new version has been submitted, the details also show the status of its latest submission and the version number.

Updating a skill (new version)

To release a new version of a text instruction skill, open its details and, at the bottom of “Overview”, click “Update” if that action is available to you. Change the description or text and save. Iskra assigns the new version number itself; the version goes through the same moderation as the first submission. If the skill changed while the form was open, Iskra won’t overwrite the newer text: close the form, open it again and redo the edit on the current version. For a skill uploaded as a finished bundle — a program skill or a text skill with resources — upload the descriptor file with an increased version and an updated source archive.

All versions of one skill take up a single slot in the skill count limit. You can release a new version of an existing skill even when the limit is reached; a new skill needs a free slot. Cancelled and rejected submissions don’t count toward this limit. If all that remains of a skill is such history, resubmitting it will need a free slot. The separate limit on the number of unfinished submissions still applies to updates.

Withdrawing a published skill

If you no longer need a skill and have permission to delete it, open its details, find the separate “Delete skill” action at the bottom of “Overview” and confirm. The skill stops being available in chats — this usually takes a few seconds.

Access keys (secrets)

Some skills need keys for external services — for example, an access key for a third-party API. You don’t have to fill them in right away: you can install a skill and turn it on in any mode without keys, and configure them whenever it suits you. If the keys aren’t set yet, a warning icon with the tooltip “Needs setup” appears in the top-right corner of the card — it opens the settings of that particular skill. Until the keys are filled in, calling the skill in chat may fail — Iskra then explains what is missing and offers to open the same settings form.

You can manage keys that are already set in the “Skill settings” window — or on the skill page, on the “Settings” tab: the “Secrets” section shows which keys

are filled in, and that is also where you replace a value with a new one or delete it. Entered values are hidden and never shown back — to replace a key, just enter a new one. The list shows only the keys you can fill in; if all of the skill’s keys are set at the organization or platform level, a single line saying that an administrator handles the setup appears instead of the list.

In your own personal skill, a secret can’t be declared with the org scope: such a submission or new version will be rejected. Remove scope — Iskra will choose the personal profile — or specify scope: profile ; if you need a shared secret, submit the skill on behalf of the organization. If a previously published personal skill shows this error, open “Update” and release a fixed version — there’s no need to contact the organization administrator.

Settings groups (connections)

Some skills need not just a key but a whole connection to an external service: a server address, a login, a couple of toggles such as “allow writing” — along with a password or token. In the “Skill settings” window, such a skill shows not a single hidden key but a settings group with a clear name (for example, “1C connection”).

Where to configure — the “Connection” row. If you have a choice — your personal profile and, if you are an owner or administrator of an organization, the organization as well — the top of the window shows a “Connection” row with a list: “Personal” and “Organization”. Below the list is a hint about the selected option. For personal, it says “Only for your profile”. For organization, it is a reminder that the data you enter, including the login and password, will be shared by all members who have access to the skill, and that such a connection is usually set up by the organization’s owner or administrator; if only you need the data, choose personal. If you are neither an owner nor an administrator, the “Organization” option in the list is unavailable, and the hint explains that the organization connection is set up by the organization administrator. If an organization connection is already configured and you choose personal, a warning appears below the row: “A personal connection will fully replace the organization settings for you”. When there is nothing to choose from — you are not in an organization or the skill has only one allowed level — there is no row at all: the window shows the fields of that level right away.

Fields. Only the fields of the selected level are shown — never both levels at once:

Regular fields — address, login, checkboxes, list selections — are visible and can be edited individually, without re-entering the other fields.

Password or token — like regular keys, it is stored hidden: you can only see whether a value is set, you can’t view it. To replace it, enter a new one; to keep the current one, leave the field alone.

Next to the group name, a label shows which connection is currently in effect: “in effect: personal”, “in effect: organization” or “not configured”, with a “Delete connection” link to its right. A connection is never assembled from fields of different levels — if both are filled in, the personal one applies. Switching the level in the list starts the setup over: if the fields have unsaved edits, Iskra first warns you that they will be lost.

One button at the bottom. The whole window is saved with a single “Save” button at the bottom: regular settings groups and keys have no buttons of their own — edit as many fields as you need and click once. If the window opened when you turned the skill on, the button is called “Enable skill”: it saves the settings and turns the skill on right away. The “Settings” tab of the skill page works the same way — the “Save” button sits below the same fields.

The exception is skills that can have several connections (for example, a mail skill with several mailboxes). There, each connection is a separate card with its own “Edit” and “Delete” buttons, below the list there is “Add

connection”, and an open card has its own “Save” and “Cancel”. You don’t have to click them: the button at the bottom of the window saves the open card too.

Leaving without saving. The “Cancel” button and the cross close the window without saving anything you haven’t saved. A special case is the window that opens on its own right after you install a skill from the catalog: there, while nothing has been saved, “Cancel” and the cross also cancel the installation itself; to keep the skill installed and come back to the setup later, use the “Set up later” button (for details, see “Skills with secrets” in the Skill Catalog article).

If such a group is required for the skill to work, you can still turn the skill on (“On command” or “Auto”) without a connection — but until the connection is configured at least at one level available to you, calling the skill in chat won’t work: Iskra explains what needs to be set up and opens the same settings window.

Examples of skills with settings groups are the 1C connection and the AmoCRM skill.

Connection knowledge map

Skills that integrate with an external system (for example, the 1C connection, the AmoCRM skill, the “Bitrix24” skill) have a “Knowledge map” tab in the skill details window — accumulated knowledge about your system: which objects and fields it has, what they mean, and verified sample queries against them. The map isn’t static — Iskra adds to it as it works, including remembering a working approach it found after an error so that it doesn’t repeat the error.

On the left is a tree of the connection’s objects by category; on the right is the card of the selected object: “Purpose”, “Fields” (what each field means), “Notes” and “Call examples”. Text fields can be edited by hand.

If there is no map yet, click “Explore” — Iskra studies the connection and fills in the map itself. A map that is already built can be refreshed with the “Rebuild map” button or deleted with the “Delete” button.

Overview

If you made a personal skill that your colleagues would find useful, you can transfer it to the organization: the skill becomes shared, and all employees can work with it. The option appears only when you work in an organization, and only for your own personal skill — a skill installed from the marketplace, a built-in skill or one that is already shared can’t be transferred.

The button is on the skill details page (open it with the “Details” button in the Skills section), in the bottom action panel:

if you are the organization’s owner or administrator — “Transfer to organization”: the skill moves over immediately;

if you are a regular employee — “Request transfer”: a request is sent, and the organization’s owner or administrator reviews it. Until the request is approved, the skill remains yours.

Before the action, Iskra shows a window describing the consequences. Read it in full: the transfer is irreversible, and the skill can’t be turned back into a personal one.

What happens to the skill

The skill becomes the organization’s property. It appears for employees — usually within a minute — and they see it alongside the organization’s other skills.

You lose the right to change it. After the transfer, only the organization’s owner or an administrator can update or delete the skill — even if you wrote it. A regular employee sees exactly that on the shared skill’s card: “Only an organization admin can change it”.

The skill gets a new name. The title you see in the list stays the same, but the skill’s internal name changes from your personal one to the organization’s — the old one no longer exists. If you referred to the skill by that name (for example, in an API integration), update the reference.

The current version of the skill is moved. It isn’t rebuilt and doesn’t go through review again — it is exactly the skill you had. Earlier (draft) versions aren’t kept.

The freed-up name is available again: after the transfer, you can create a new personal skill with the same name.

What does NOT move

Everything the skill accumulated while it was personal is erased during the transfer — the organization doesn’t inherit your data:

settings and access keys (secrets) — they have to be set again. Regular keys stay personal: each employee sets up their own. If the author declared an organization-scoped settings group for the skill, that group is what an administrator fills in once for the whole organization — but they won’t enter your personal keys for you;

the activation modes (“Off”, “On command”, “Auto”) that you or your colleagues chose for yourselves;

consents for the skill to access data;

the connection knowledge map — it will have to be built again;

scenarios you saved on top of the skill. Scenarios the author attached to the skill move along with it;

the background skill’s schedule. This matters: if the skill ran in the background on a schedule, background runs won’t resume on their own after the transfer. You need to set them up again on the shared skill — otherwise the regular work will quietly stop;

data the skill accumulated while working — its records and files. For a tracker, log or collector skill, that is all of its content: the organization receives the skill empty. If you need the data, export it before the transfer.

The run log history stays under the skill’s old name; for the shared skill, it starts from scratch.

When a transfer isn’t possible

Iskra explains the reason right in its reply. Most often it is one of these:

the organization already has a skill with that name — an administrator can resolve the conflict: delete the existing shared skill or reject the request;

the skill is listed on the marketplace or has a pending publication request — remove it from the marketplace first;

the skill is being edited or reviewed — wait until that finishes and try again;

a transfer of this skill is already in progress — wait until it closes (usually a few minutes) and decide on the request again;

this version’s manifest no longer matches the registry — retrying won’t help: publish a new version of the skill and transfer it;

you are not a member of the organization, or the organization is blocked.

Sometimes Iskra replies that the skill has been transferred but part of the move is still being finished automatically. That is a success, not an error: the skill already belongs to the organization, and the platform completes the rest by itself within a few minutes — you don’t need to click anything again. Some of the shared skill’s settings will not appear right away.

The request: submitting, status, withdrawal

After you submit, the skill row shows the “Transfer request — under review” status and a “Withdraw” button. The status persists after a page reload. A request can be withdrawn only in the Skills section — this can’t be done from chat.

The same row can also show two other states:

“Transfer in progress” — a decision has been made, but the outcome isn’t known yet: Iskra is finishing the move in the background. You don’t need to click again; the outcome will appear right here;

“The transfer did not go through” — the move was closed unsuccessfully, and the skill remains yours. The transfer button next to it stays: you can try again.

The request records the version of the skill that the administrator will see. If you release a new version before the decision, the request can no longer be approved: the administrator gets a refusal saying “the skill has changed since the request”, and the request itself is cancelled automatically — submit a new one. The request is also cancelled if, by the time of the decision, you are no longer a member of that organization.

You get a notification about the decision on your request; if it was rejected, the notification shows the reason the administrator wrote.

Reviewing requests (organization owner and administrator)

Employee requests are reviewed only in the Skills section: a “Skill transfer requests” block appears above the skill list. The link in the notification opens it directly.

A request row shows the skill title, the version, the submission date, the author and the name the skill will have in the organization. The “Show skill” button expands the description and full text of the version recorded in the request — so the decision is based on it rather than made blindly. Next to the text there is always a collapsed “Manifest: permissions, access, schedule” block: it shows what the skill can do — which capabilities it requests, which networks it reaches, which access keys it declares and whether it runs on a schedule.

The same block also lists the organization’s unfinished and failed transfers — with a status instead of decision buttons: “Transfer in progress” and “The transfer did not go through” with the reason. There is nothing to decide on them, but you need to know about a stuck transfer: the operation is irreversible.

If there is nothing to show at all, Iskra warns you directly: ask the author or reject the request. A request withdrawn by its author can no longer be opened — the offer has been taken back.

Then there are two actions:

“Transfer to organization” — with an extra confirmation: the author loses the right to change the skill, and it can’t be turned back into a personal one;

“Reject” — a reason is required: it is sent to the author in a notification.

If another administrator has already decided on the request, Iskra tells you so and offers to refresh the list.

Transferring from chat

You can do the same without opening the section: tell Iskra “share this skill with colleagues”, “make this skill shared” or “transfer the skill to the organization”. Before the transfer, Iskra shows a form with the same consequences and a “Transfer to organization” button; to back out, use the “Cancel” button. The outcome depends on your role: for an owner or administrator, the skill moves over immediately; for a regular employee, a request is sent. Withdrawing a request and reviewing other people’s requests can’t be done from chat — only in the Skills section.

Overview

For owners of websites and apps that show YAN ads: revenue, impressions and clicks for your sites and apps, totals and daily trends. Requires a personal statistics API token for a registered YAN account. Advertiser spend and campaigns belong to Yandex Direct; site indexing and search queries belong to the Yandex Webmaster skill. Setup and how the services differ: the YAN guide.

Overview

For Yandex Direct advertisers: campaigns with their status and key goals, spend excluding VAT, impressions, clicks, CTR, average cost per click, conversions by Metrica goals and cost per lead for the period — totals, daily trends and a breakdown by campaign. Requires a token from your own OAuth app with an approved request for Direct API access; the skill only reads statistics. Setup and how the services differ: the Direct guide.

Overview

Synced folders from your computer appear in the unified “Memory” section along with personal folders, project folders and system-managed chat folders. The sync itself is still set up in the Iskra desktop app.

Memory in the web version

Open the “Memory” navigation item. A synced folder shows its source; click the folder to open its contents.

Inside a folder, files and subfolders are shown as a list. You can drill down into subfolders, and a navigation trail (“breadcrumbs”) at the top lets you go back to any level. Each file shows its size, and you can download it.

To use the folder in a chat, connect it via “@ Folders” — a synced folder can’t become a working folder (see “Chat working folder” in the “Chat” section). A chat has either its own system-managed folder or a Memory folder chosen in advance — its contents are shown in the “Chat files” area of the prompt panel: that is where the files created in the chat itself are kept, not the contents of a connected synced folder.

You can’t set up or change the sync itself in the web version — that is done in the Iskra desktop app. The web version shows the contents of synced folders in Memory.

Placing files from a synced folder on a board

A placement isn’t created in Memory: the “Publish on a board” item lives in the file menu of the chat files panel. Memory lists — including the contents of synced folders — show existing placements but don’t create them. A placed file has a short placement mark in lists, and in roomier places — the preview and the file card — “Showing the version from …” and “Source changed …” appear next to it: this tells you whether the latest edit has reached readers.

An edit made on your computer comes back to Iskra through regular sync and becomes a new version of the file — just like an edit made in Iskra itself. The tab mode decides what readers see after that; the mark on the file shows the outcome of that decision.

A deleted source, including one moved to the Iskra trash, is not served to readers, even if the tab version was pinned; such a file has no placement mark either. Restoring the file doesn’t bring the placement back by itself — that takes a separate “Restore the placement” action. For details on boards, audiences and version modes, see the “Boards” section.

Sync via the desktop app

The Iskra desktop app can sync a folder on your computer with the platform. Sync is two-way: changes in the local folder show up in Iskra, and changes made through the platform come back to the folder on your computer. This keeps files up to date both on your computer and in Iskra.

The “Drive” page opens via the “Computer sync” action in the “Memory” section (desktop app only); if unified Memory is turned off, the “Drive” item is right in the sidebar. The page offers:

Attach folder — choose a local folder and start syncing it.

List of attached folders — each folder shows its state; you can pause sync temporarily and resume it, open the folder on your computer or browse its contents in the app.

Detach folder — stop syncing. When detaching, you can choose whether to keep the local files on your computer.

Server folders — previously synced folders can be downloaded and synced to the current computer. Regular Memory folders don’t appear in this list.

Sync connects automatically: just sign in to the app with your account; there’s no separate sync password to enter.

If two versions of the same file have changed differently, a conflict occurs. The app highlights such cases and asks you to choose which version to keep.

The “file deleted in Iskra” conflict ( deleted_path_conflict ) means that a file with the same name is in the Iskra trash. Sync doesn’t rename the existing local file automatically: restore or rename the file in the trash, or rename the local file, and then sync again.

From Drive to Memory, a chat or a project

An attached folder is already in Memory: you don’t need to add it there separately or upload a copy of the files. Work under the same profile in which you attached the folder.

1. In the desktop app, open “Memory” → “Computer sync” and

click the name of the attached folder. A view of its contents opens in Drive.

2. Choose “Open in Memory” to go to its files, or “Connect to

a new chat” to start a chat with this folder as a context source. Both actions apply to the whole folder, even if one of its subfolders is open in Drive. In a new chat, the connection is saved once the first reply is complete. Wait for it, check the folder in “@ Folders” and then ask questions about its materials.

3. For a chat that has already started, open that chat and connect the folder via “@ Folders”.

You can add a second folder the same way.

For a project, open the main “Memory” list, find the folder and in its “More” menu choose “Connect to project…”, then the project of yours that you need. If the folder is hidden by a saved filter, choose “All” or “Connected”.

Connecting gives the chat or project access to the materials. It doesn’t make the synced folder the chat’s working folder and doesn’t automatically move all of the chat’s files into it.

From Memory to your computer: what is supported

A previously synced folder of your profile can be downloaded to this computer if it isn’t attached on it yet. Sign in to the desktop app under the same profile, open “Memory” → “Computer sync” and find the folder among the server folders. Click “Sync to disk”: the app creates a local folder for it inside ~/Iskra/Drive/ . Folders already attached on this computer are in the upper list of attached folders.

A regular personal Memory folder or a system-managed project or chat folder can’t be made a synced folder from this screen yet. There is also no operation here for linking an existing Memory folder to a chosen existing folder on your computer. “Attach folder” creates a separate synced folder rather than merging it with a Memory folder of the same name.

Downloading an individual file from Memory saves a copy to your computer; it doesn’t turn on continuous sync.

Operating system support

Sync works reliably on macOS.

Beta: sync on Windows and Linux is in beta and may be unstable. Support for these systems is being actively improved. If you use Windows or Linux, the desktop app shows a corresponding warning on the “Drive” page.

Overview

An API key gives access to Iskra from scripts, bots and external systems. You can get catalogs, choose skills and specialists, connect Memory folders and collections, prepare a conversation and run a task. The key is created in Settings → API Keys; the “API docs” window shows examples with the address of your installation.

All calls below use Authorization: Bearer <token> . For a JSON body, add

curl -F , the content type and form boundary are set by curl itself).

Capability Method Key permissions (scope)

Specialist catalog and plan preview

Collections and their members s , GET memory:read

Capability Method Key permissions (scope)

Create persistent folders, upload a file memory:read + memory:write

Download a persistent file memory:read

Prepare a conversation without starting a task POST /api/v1/conversations chat:run ; Memory permissions for the selected targets

Read and change the context — selecting a workspace and the write mode also require

Background task, polling, events, cancellation /api/v1/runs/{run_id} , GET

conversation files /api/v1/chat/{id}/files/{p respectively

or model tions , GET also llm:raw

App publishing apps:publish

Web search requires web_search , running skills requires skills:run , and passing attachments requires files:write ; Memory context adds its own permissions on top of the run and file permissions.

The usual order: catalogs → choose an available bundle → POST /api/v1/conversations

iskra_exec → download the result. In memory_refs the target is set via locator:

secondary_specialist_id , profile_id (the model profile, not the user profile), complexity ( simple , normal , hard ), antonym . A manual skill set includes all

configures runs of skills that are already available; installing and configuring a skill is done in the interface.

The memory:read and memory:write permissions are selected explicitly when you create a key; writing requires both. Key permissions are limited by your profile’s current access. The working folder can only be a personal Memory root with write access, chosen before the first message or upload. After memory_refs are removed, the conversation still requires memory:read to continue and to access files and results: the history may have retained materials that were read. Memory files are persistent; the key’s expiry and the conversation TTL are set separately. For a detailed runnable example, see “Full API scenario”.

Overview

1. Open Settings → API Keys.

2. Click Create key.

3. Enter a name — it only serves to tell keys apart in the list

(for example, “CI bot” or “Export script”).

4. Choose a validity period: 7, 30, 90, 180 or 365 days. The default period is 30

days.

5. Click Create.

The expiration date is shown after creation and in the key list. Choosing a new period does not change the dates of keys created earlier. When the period ends, the key stops working; to keep the integration running, create a new one. If the creation settings are temporarily unavailable, you can create a key with the server default period of 30 days; choosing a different period is not available in this state.

Before creation, the window shows a list of access scopes with checkboxes. The standard permissions are already checked: chat, web search, running skills, and writing and downloading conversation files — you can uncheck the ones the key does not need (for example, keep only chat). Sensitive permissions are unchecked by default and are granted only explicitly. The final set of permissions is visible before the key is created; treat the token like a password.

Sensitive permissions in the list:

“Reading Memory” (memory:read) — lets the integration view the folders, files and collections available to your profile, connect materials for reading and get the results of working with them.

“Modifying Memory” (memory:write) — lets the integration create folders, upload materials and allow writing to connected Memory. It is granted together with the read permission; the write permission alone does not include reading automatically.

“Runs on behalf of organization members” (run:delegate) — available only for organization profile keys. Check it only if the key must run requests on behalf of other employees (for example, an integration that processes documents for a specific employee): such a key can act on behalf of any colleague in the organization.

“Raw model access” (llm:raw) — opens the OpenAI-compatible mode for calling the model directly with your own tools. The permission is added to the key’s other checked permissions; in the direct-call mode itself, Iskra’s agent features and skills do not apply.

“Publishing apps” (apps:publish) — needed by the key that the iskra-app command-line tool uses to create an app and publish its versions. The permission works only in your profile: you cannot use it to publish an app to someone else’s profile, even

if the key has the permission for runs on behalf of employees. How apps themselves work is described in the “Apps” chapter.

Reading boards (boards:read) — gives the integration access to the material of the boards your profile manages: the list of boards, the composition of the displayed version, its content and downloads.

Preparing board placement (boards:publish) — lets the integration prepare the placement of a file on a board. The key does not perform the placement itself: a person confirms the prepared action.

Both scopes — boards:read and boards:publish — are granted only explicitly and are not part of the key’s standard permission set: a board shows finished documents to people outside the chat, so a key gets access to it only by a separate decision. Details are in the “Boards via API key” section below.

The list shows only the scopes allowed on your platform: if a permission is missing from the window, it is not available in this installation — contact your administrator.

The token is shown once after creation. Copy it immediately — you cannot see the token again, because the platform stores only its hash. If you lose the token, revoke the key and create a new one.

Overview

setup_state. iskra_exec.skills takes bundles, not tool names. requires_setup shows whether setup is needed; setup_state can be ready, missing, invalid or unknown. A row in the catalog does not by itself mean the skill is ready to run. If a selected or mandatory skill is unavailable, first finish its setup in the interface and fetch the catalog again.

plan/defaults returns plan, pins, version and options, including the available levels, mandatory_bundles and specialists. This is a preview before the conversation is created that helps you compose the request.

iskra_exec field How to choose

An allowed model profile from the available options; it differs from the user profile the key is bound to

complexity is chosen explicitly; usage depends on the

actual model

Primary specialist: sets the full working style

Additional subject-matter expertise

secondary_specialist_id together with the primary one; at most two different

roles; “General” is not used in a pair

A manual array of bundles that includes all

skills mandatory skills; each selected skill is

available and ready

iskra_exec field How to choose

true enables Antonym mode; an explicit false disables it unless the mode is locked by a platform or organization policy

A missing field leaves the decision to automatic selection. For skills, a missing field or null keeps the usual selection; skills: [] means a manual empty set and is allowed only when there are no mandatory skills. Otherwise, 403

one response with primary and additional expertise. All iskra_exec values apply to a single request in native, async and OpenAI agent; pass them again for the next turn if you need them. They do not change the conversation’s persistent settings. Without an override, the conversation uses the saved automatic specialist selection; the API complexity without an override is normal.

Overview

The key is explicitly granted memory:read for reading and, in addition, memory:write for writing. The files:read / files:write permissions apply to conversation files and do not open Memory by themselves. A key that has already been created does not get new permissions automatically.

<uuid> for the root contents, &folder_id=<uuid> for a subfolder. Available filters: q,

Use the IDs and locator from the catalog response.

Memory write or download Body / parameters

201, a root with id

the response is the file bytes

For upload, path defaults to the file name; conflict_policy: create rejects a name collision, rename picks a free name. Files and folders are persistent: the conversation TTL does not delete them. Access permissions, quotas and upload restrictions still apply.

The Bearer API provides read access to collections and their membership. Creating collections, editing their membership and deleting them is done in the interface; there are no bearer CRUD routes for collections. The resource write operations in the table create roots and subfolders or upload files.

Overview

POST /api/v1/conversations with the chat:run scope creates a conversation without a message and without starting a task. The minimal body is {}; Memory setup:

Each of up to 100 references sets exactly one target: a locator with directory_id and

or write. Targets must not repeat; the working root is already connected as the workspace and

references. On error, the context is not created partially; a repeated POST creates a new conversation.

The workspace can be selected or removed only before the first message or upload: it is a personal root with write permission, not a subfolder, a collection or an organization root. Replacing an already selected folder returns 409; repeating the same selection is idempotent. Without a selection, files are stored in the conversation’s temporary folder. In a selected root, results are saved to Memory and survive deletion of the conversation.

Context of an existing

Body Permissions beyond chat:run

conversation

One target, as in the array memory:read, for write

to remove a write reference

Access to resources is checked on every request. A write connection requires write permission and permission for the agent to modify the material. For a collection, before a conversation is created, a reference is added or a PATCH to write is made, every current member is checked: its

write permission and permission for agent writes. A saved write mode requires both Memory permissions at run time, even if the collection is empty or access was later narrowed to reading.

A conversation that has been given access to Memory at least once keeps requiring memory:read after all references and the workspace are removed. This applies to native, OpenAI agent, async, file upload/download, result polling and SSE. Deleting the conversation itself remains available to the owner with the chat:run scope. During an active turn, the context is locked (409). If the context changed between queuing an async run and its start, the run gets failed / context_changed: send a new request with the current context.

Overview

The key is bound to the user’s active profile selected at creation (personal or organization). Switching profiles changes the list of keys. The key itself does not connect a project or knowledge base: materials are selected explicitly through the conversation context. iskra_exec.profile_id is a separate model profile identifier that does not change the owner of the key or the conversation.

An organization key with an explicitly granted run:delegate can act on behalf of an employee of that organization: X-Iskra-Run-As: <employee profile_id>. Usage and access to materials are attributed to the selected employee. Pass the same header throughout the whole chain of upload, run, polling, SSE, download and cancellation; without it, the request addresses the key owner’s profile and may get 404. Apps publishing works only with your own profile and rejects delegation.

Overview

and, if there are results, files. A conversation created via POST /api/v1/conversations is used the same way.

Pass the conversation_id from the previous response and a new message. This is the ID of a separate API conversation, not a setting that connects the key to a project or knowledge base. One conversation is used sequentially in native, async and OpenAI agent. A concurrent turn gets 409.

The Idempotency-Key header in native and OpenAI agent deduplicates only conversation creation: a repeated request can run another turn. For background POST /api/v1/runs, a retry with the same key returns the previous run_id and does not start a second execution.

Request fields

native / async field Meaning

message Required message text

A prepared or continued conversation; without it, a new one is created

instructions Additional instructions for this turn

files {id} from a multipart upload; async accepts

only {id}

native / async field Meaning

Conversation lifetime: 3600 seconds by default, 86400 maximum

of the requested form

true adds the available reasoning to reasoning; off by default

{...}} for structured output

Fields for choosing the model, complexity, skills,

iskra_exec two specialists and Antonym are described

above; they apply to one request

usage shows token consumption approximately and is not a billing calculation. completed means the response is ready. For interaction_required, show the person

conversation_id and the values in policy.inputs. The response also contains interaction.tool_call_id. An empty confirmation form is completed in the interface: policy.inputs passes field values but does not give consent on the person’s behalf.

Overview

passes nothing: Iskra receives a file as content or as a reference to a prior upload.

In files, each element specifies exactly one option:

{name, mime, content_base64} — bytes in base64 for native sync;

{id} — the version ID from the POST /api/v1/chat/files response. This is the only source of an input ID. The files array of a turn result contains only {name, download_url} and does not provide such an ID.

Upload requires files:write and accepts the multipart fields conversation_id, an optional ttl_seconds and a repeatable file field. Without conversation_id, a new conversation is created.

workspace, both Memory permissions are also required; after the conversation has accessed Memory in any way

The ID is valid only in this conversation and only once: reattaching it to another message, or using someone else’s or a nonexistent version, returns 404 invalid_file. Upload the file again to use it again. Both options at once or neither: 400 invalid_request; async with inline: 400 use_file_upload. A separate Office processing limit may reject an already uploaded file before the turn (413 file_size_limit): reduce or split the file.

An attachment passes a specific file into the turn. To work with a Memory folder or collection, connect it via memory_refs.

How to download a created file via the API

Files created by the agent are listed in files: [{name, download_url}]. Download the actual download_url with a Bearer key and files:read, for example GET /api/v1/chat/{id}/files/report.md. For a conversation with Memory, memory:read and current access to the selected working folder are also required. Through the Memory API, a persistent file is downloaded by file.id from the catalog, not by the upload version ID.

Overview

POST /api/v1/runs accepts native fields (attachments only by id) and immediately returns

a fixed run deadline, not the conversation TTL.

next_after. While the status is queued or running, keep polling; in the next request, pass next_after, not the number of events read. Final states: completed,

the machine-readable failure code is in error.code.

same feed via SSE. The Bearer header is still required: the browser EventSource cannot set it by itself — use a client that supports headers, for example streaming

report progress: content contains the text length, tool the tool name; the response itself is fetched by polling the result. The terminal event closes the stream.

DELETE /api/v1/runs/{run_id} cancels a pending or active run; for a finished run, 409 already_finished is returned. After a service restart, an interrupted run may remain running until the deadline check and then ends as failed / interrupted; retry based on the terminal outcome.

Overview

the agent model is iskra or iskra-agent. GET /api/openai/v1/models returns the available names; the surface is enabled by an administrator, otherwise 404.

Skills, specialists and connected Memory work in agent mode. Pass

additional fields there is extra_body. With conversation_id, the server keeps the history and, from the new messages, uses the last user message and the system / developer instructions. Without an ID, each request is standalone. stream: true is supported; the response contains

agent mode are rejected: Iskra uses its own tools.

powerful and requires both chat:run and an explicitly granted llm:raw. It is available when enabled by an administrator. This is a call to the model with the client’s history and tools: skills, specialists, Memory context and Iskra’s server-side conversation do not apply here; conversation_id, policy and iskra_exec belong to agent mode. The model returns tool_calls, and the client executes them. Raw supports

response_format with the value text or json_object; for JSON Schema, use native or async.

Overview

answer, and the object matching the schema arrives in output. An invalid schema is rejected before the run (400 invalid_schema). If the response could not be coerced to the schema, native returns 422

interaction_required and errors, schema coercion is not performed. In OpenAI agent

Overview

The key’s validity period relates to authorization: 30 days by default, 365 days maximum. The API conversation TTL is separate: ttl_seconds, 1 hour (3600 seconds) by default, 24 hours (86400 seconds) maximum; a turn or upload extends it. These conversations are separate from regular interface chats. Deleting or expiring a conversation closes its file URLs; persistent files in the selected folder and connected materials remain available through Memory.

A synchronous turn is limited to 90 seconds, a background run to 15 minutes from creation. When retrying a request, take the Idempotency-Key rules above into account.

HTTP What to do

Fix the body, fields, IDs or schema according to the machine-readable error code

The key is invalid, expired or revoked: create a valid key

Check the scope, access and policy: a missing scope can be

route

The resource does not exist, belongs to someone else, is unavailable for

404 reading, or the conversation has expired; it is also

possible that the API is disabled

The conversation is busy, it is too late to change the selected context, or the run has already finished

410 The resource is sealed and the operation is unavailable

The file upload or processing limit has been exceeded

the original response and check the schema

The request rate is limited or the quota is exhausted

HTTP What to do

Internal error, empty_response or

before retrying

A dependency/catalog/lock is temporarily

503 unavailable, or maintenance is in progress; with

504 The turn timed out

The error format depends on the route: native returns error.code and error.message, and context operations may return problem JSON with code, detail and params. Check the HTTP status and the machine-readable code, not the exact text. In OpenAI SSE, an error after the stream has started arrives as an error frame, with HTTP 200 already sent; async errors are read from the terminal result.

Overview

An example for Bash, curl and jq: catalog → a selected ready-made skill → sources and a Memory working folder → conversation → task → persistent result. Create a key with

your installation address and the token via environment variables.

Choose the bundle and id that fit the task from the printed catalogs. The next block requests them explicitly and checks that the selected skill and all mandatory skills are available. If the check fails, finish the setup in the interface and fetch the catalog again. These routes do not install skills.

If you need a background run, instead of the last call send turn.json to POST /api/v1/runs, wait for a terminal status via GET and save its .result to result.json. For OpenAI agent, convert the request: jq

/api/openai/v1/chat/completions. Each variant you send creates a separate turn.

Download the link from the actual response: the model may choose a different name or ask for clarification. If there are no files, first examine status, answer and interaction; the task itself does not guarantee that an artifact is created.

For a read-only integration, grant memory:read without memory:write, connect

unset. The results will be temporary conversation files.

Overview

The list shows all keys of the current profile: name, prefix (isk_…), permissions, creation date, last used date and validity period. The token itself is not shown.

To disable a key, click Revoke and confirm. The action is irreversible — the key stops working immediately. Revoke a key if the token leaks or when the integration is no longer needed.

Overview

/api/v1/keys/scopes require a user session. Creating keys, listing them, the permissions catalog and revoking keys are done through the interface; a Bearer key cannot manage them. The token is shown only once.

With an explicitly granted apps:publish, the iskra-app CLI creates an app via POST

its accepted publication via GET /api/v1/apps/{id}?version={version}. Publishing applies only to the key owner’s profile. App access on behalf of its viewer works separately: see “Apps”.

Overview

With an explicitly granted boards:read, the key reads the boards of the profile it belongs to: GET /api/v1/boards — the list of boards managed by this profile, GET /api/v1/boards/{id} — the board’s composition and displayed version, GET

file of the displayed version. A board where the profile is only a recipient can be read by its ID, but it does not appear in the list. The key does not change the audience, the secret link, invitations, copying or ownership transfer — these are actions a person takes in the interface.

A tab’s download address is not the link a board reader receives: it works only with your key and the boards:read scope. A board opens to a reader through its audience — the organization, an organization group, a named invitation or

a secret link — so there is no point in giving the API address to a reader: it does not open without the key.

With an explicitly granted boards:publish, the key only prepares the placement: POST /api/v1/board-publication-intents returns the exact form of the future action — the board, its

shows the preparation status and the saved result. There is no confirmation by key: a person opens the prepared action in Iskra on the /boards/publication-intents/{id} page and confirms it with their own session; the preparation waits for them for a limited time. The boards:read scope does not create preparations, and boards:publish does not open board reading: these are different permissions of the same key.

Overview

The token is stored as a hash — the platform cannot recover it.

Keep the token secret (environment variables, a secrets manager), and do not commit it to a repository.

If it leaks, revoke the key immediately and create a new one.

A key acts strictly within its profile and the permissions granted to it.

Overview

You can use Iskra’s API SDK for integration. It supports chat, background tasks, files, skills, specialists and Memory; the Python client is available in synchronous and asynchronous versions. Installation instructions from the public source and examples are in the README of each package. Pass the key and your installation address explicitly. For photos and long-running jobs, use file upload and an asynchronous run.

Overview

If an action fails, Iskra shows a short message and a hint about what to do next. Technical details of an internal error are not displayed on screen: they are saved in the platform’s diagnostics so that support can find the cause without exposing unnecessary details to the user.

What the system may suggest

Hint What to do

Fix the form fields, the selected file,

Check the entered data the link or other input, and submit

the action again.

Repeat the same action right away. This kind of

Try again error is usually caused by a temporary request

failure.

Wait a few minutes and repeat the action. This may be a temporary

Try again later

overload or unavailability of an external service.

Hint What to do

The session has expired or the action requires

Sign in to your account authorization. Sign in again and repeat the

action.

The action is restricted by

Contact your administrator organization settings, access permissions or

platform settings.

Repeating the same request will not change anything: you need to do something else first —

Do not retry without changes

for example, remove a skill from the showcase or submit a new request.

The request covers too much data:

Narrow the request

add a filter, a period or more details.

Iskra needs your decision: confirm

Clarify the action the action or choose one of the

options offered.

A plan limit has been reached, or the feature

Upgrade to a suitable plan

is not available on the current plan.

The error requires diagnostics by the

Contact support

support team.

If a request ID (request_id) is shown on screen, save it. When contacting support or your administrator, include:

what you were doing before the error;

the approximate time of the error;

the request ID, if it is shown;

the page or section where the error occurred.

How to report an error in chat

If a chat response ended with an error, the error card has a “Report a problem” button. It opens the regular feedback form — the same one used to rate a response — and sends the message to the team.

You do not need to copy diagnostics manually: the error type, the request ID and a link to the chat are attached automatically. A comment is optional, but a few words about what you were doing before the error speed up the investigation. After you send it, a confirmation appears that the message has been received.

The same button is also inside the “Error details” window if you have already opened the details. The request text, file contents and the model’s response are not included in the report.

When you can simply retry

Retry the action if the hint explicitly suggests “Try again” or “Try again later”. For file uploads, payments, access permission changes and other sensitive operations, first check the result on the page: sometimes the operation may have completed even though the interface did not receive the response in time.

The message “The AI service rejected the request” means that the response service did not accept the request. It does not point to an error in your text. You can try again;

if the refusal repeats, click “Report a problem” in the card. The type of refusal is saved in the chat history so that the team can find the cause.

If a chat message is not accepted

If the server rejected a message before the response started, Iskra keeps it above the input field with the label “Message not accepted”. This pill is not resent automatically and stays in this tab after a page reload. Subsequent messages wait so that their order is not broken.

Check the error hint, then choose an action on the pill:

Edit — change the text before a new attempt;

Retry sending — explicitly send the message again;

Remove from queue — delete the message and move on to the next ones.

If the error requires a different action — for example, signing in again or changing the file, plan or access permissions — do that first and only then retry sending. When Iskra asks you to sign in again, the message waits for your return: after you sign in to the same profile, it appears above the input field again with the same label.

A separate case is an attachment that failed the size check: such a message stays in the input field itself together with the attached file instead of turning into a pill. Replace the file with a smaller version and send it again. Changing your plan helps only with a plan limit; it does not raise the technical processing limit. When you select a new Office file that is too large, Iskra shows the reason in advance, before adding the attachment and transferring its contents.

If a response is blocked during execution

For a blocked response, Iskra may offer a “Retry from a clean step” button. It repeats the same request once without the unfinished response. The previous history and the working folder remain available.

Check the result before clicking: the new run may repeat file changes in the working folder and actions in connected services. It does not cancel or roll them back.

If Iskra has managed to complete some of the steps, “Continue” remains next to it: it completes the response that was started, while “Retry from a clean step” starts the request over.

The button is available only in the current error card and is not restored after a page refresh. There is only one “clean step” attempt: if the response is blocked again, send a new message or rephrase the request. If the new run is interrupted for a technical reason — for example, the response service is temporarily unavailable — the regular request retry remains available.

After the connection is restored

Once the connection is restored, Iskra re-reads on its own the updates that may have been missed during the outage: the chat list, notifications, the open chat and usage figures. The page is not reloaded, and unsaved input stays on screen.

The only case when Iskra reloads the page by itself is when a new version of the app was released during the outage. A banner with a countdown appears: the page reloads in a few seconds. Click “Later” to finish your action first — the banner stays over the top edge of the screen without shifting the interface or covering the input field (while the connection is down, the reconnection banner temporarily replaces it), and you can then update with the “Reload now” button. Typed text, collected quotes and the selected context — the working folder, Memory folders, scenario and tools — are saved before the update and restored after it. If the browser could not

save the input, or local files are still attached to the message, the reload is cancelled and the banner shows the reason: first finish sending or remove the attachments. The countdown runs only while the tab is open and the connection is restored — the page does not reload in the background.

If Iskra could not load part of the interface, a new version is not confirmed and the browser is not the issue, a separate neutral message appears. It does not mean that an update has necessarily been released: first check your connection, then retry the action or click “Reload”. In this case the page does not reload automatically. Before a manual reload, Iskra applies the same input-saving rules and does not proceed if local attachments could be lost.

Another case is when part of the app does not open because of the browser version. Then the message says exactly that: this browser version can’t run the app. There is intentionally no reload button next to it — refreshing the page would lead to exactly the same result, because the problem is not with loading. What helps is updating the browser or the operating system: on a Mac, that is “Software Update” in System Settings; on Windows and Android, update the browser itself. The Iskra desktop app uses the system’s browser engine, so a system update fixes it too. After that, the app opens as usual.

Slow connection: gray placeholders instead of a blank screen

When the connection is slow, the interface cannot respond instantly: the browser needs to load part of the app. So that this does not look like a freeze, Iskra shows gray placeholders — rectangles in place of the upcoming content. The menu item you clicked or the chat you selected is highlighted immediately, even before the section has opened, so the highlight shows that your click worked. Placeholders appear only where the wait is actually noticeable; on a fast connection you will not see them.

The same happens when you first open Iskra: while the app is loading, placeholders are shown in place of the sidebar and the content, and their shape depends on what will open — a chat or a section. On a slow connection, the first launch takes tens of seconds, and that is normal. If loading takes far too long, the placeholders are replaced by the line “Loading is taking longer than usual” with a “Refresh” button.

Mobile app: “Couldn’t open Iskra”

After you sign in to the mobile app, the main screen opens under a branded splash screen with the logo: it stays until the interface confirms that it is ready to accept input. If this does not happen within the allotted time (no network, the server is unavailable, the built-in browser failed), the app tries to recover once by itself and then shows the “Couldn’t open Iskra” screen with two buttons:

Retry — check the session again and open the interface; suitable after the connection is restored.

Sign in again — clear the saved session and return to the sign-in screen; use it if “Retry” does not help.

Below the buttons, a short code in the form stage/reason is shown (for example, initializingWeb/webInitTimeout). It contains no personal data — include it when contacting support, together with your phone model and the time of the error.

Sometimes the built-in browser cannot be safely restored inside the running app. Then, instead of the usual buttons, a “Restart Iskra” screen appears, asking you to close the app completely and open it again. The “Retry” and “Sign in again” buttons are intentionally absent from this screen: you can continue working after relaunching the app.

Scheduled platform update

Iskra warns you about an update in advance: a bar with a message from the administrator appears at the top — for example, “Iskra is switching to update mode; you will be able to continue in about 20 minutes”. It is visible on any screen, not only in chat, and appears by itself; you do not need to reload the page.

While the update is in progress, new messages are not accepted: the same explanation is shown in place of the input field, and an attempt to send a message returns that explanation as text rather than as an error. This is not a failure — the platform has deliberately stopped taking new tasks so as not to cut other people’s tasks off midway. Text you typed earlier is not lost and will return to the input field.

Along with the input field, the buttons that start Iskra again also stop working: “Retry” and “Continue” on an interrupted response, “Continue” and “Finish” on a scenario, “Run” in a skill chat, and saving an edited message. Each of them says why it is currently unavailable. You can still prepare your next request as usual — type text, attach files, choose a scenario: the update does not affect that.

A project’s personal chat follows the same rule: during the update, the input field does not accept a new request, and the typed text is kept until the platform resumes work.

Update mode itself does not interrupt a running task. The “Stop” button keeps working if you decide to stop the response. If problems occur, an administrator can also stop the current response; the saved chat history remains available. You can send a new request after the update.

The bar and the lock are removed automatically as soon as the platform is ready to accept requests again — also without reloading the page.

When you need to change what you are doing

If the error is related to input, your plan or permissions, retrying without changes usually will not help. Correct the data, choose another file, sign in again, change your plan or contact your organization’s administrator.

Overview

The skill catalog is a showcase of ready-made skills inside your Iskra. Skills are published to it by other platform users and by Iskra itself. Unlike system skills, which everyone sees right away, a catalog skill appears for you only after you install it — so your skill list does not get cluttered with other people’s skills.

The section is available if the skill catalog is enabled by the platform administrator. The showcase lives inside the “Skills” section (/skills), on the “Catalog” tab; old links to the separate catalog page lead there too. Your personal installations are shown on the “My skills” page in the “From catalog” section, and organization installations in the “Organization” section.

You can do all of this right in a conversation with Iskra, without opening the “Skills” or “Skill catalog” pages: ask “find a skill for working with …” — Iskra searches the showcase and shows suitable options with descriptions; “install the … skill” — it installs the selected one (for the organization if you are the owner or an administrator); “update the … skill to the latest version” — it moves the installation to the new version; “publish my … skill to the catalog” — it submits a publication request; and “remove the … skill from the showcase” or “cancel the publication request for …” — it removes the skill or withdraws the request. Iskra asks you to confirm each such action before performing it.

Overview

The showcase displays skill cards with a clear name and a short description. If the skill’s technical name differs from its name, it is shown on a separate line after a / — just like in the “My skills” list. The left side of the card shows the skill’s icon, if it has one; otherwise, a colored placeholder with the first letter of the name. The author, version and category are not shown on the compact card. To find what you need, use:

Search — by the skill’s name, technical name and description.

Categories — the showcase section headings (“Productivity”, “Integrations”, “Data”, etc.); categories without skills are not shown.

The names and short descriptions of Iskra’s built-in skills are shown in the interface language — Russian, English, French or Chinese; changing the language in settings immediately updates the cards, the showcase, the hints that appear when you type “/” in chat and the “Runs” tab. Skills published by other users and organizations keep the author’s text as is: in the interface in any language other than Russian, such a description is preceded by the “author’s text” label (tooltip: “Skill author’s text, not translated”). The names of author skills do not get this label. Tool descriptions on the skill page remain the text from the manifest.

Clicking a card opens the details: the author, the current version and the list of tools — what exactly the skill can do. The author “Iskra” means that the skill was published by the platform itself. If the skill needs keys or parameters to work, Iskra offers to configure them after installation; you can postpone the setup and come back to it later (see “Skills with secrets” below).

Overview

On the card, click the “+” icon button on the right — it installs the skill right away; in the skill details, the same action is labeled “Install”. If Iskra asks where to install the skill, choose “Only for me”. The skill appears in the “From catalog” section of the skill list and gets the “Auto” mode. You can immediately change the mode to “On command” or “Off” on the skill’s details page.

It is important to understand that installation pins the skill version. You work with exactly the version you installed — even if the author later publishes a new one. Updates do not arrive automatically: you decide when and to which version to move (see “Updates” below).

Skills with secrets

If a skill needs access keys (secrets), this does not prevent installation — it completes right away. After that, if the keys are not set yet, Iskra opens the “Skill settings” window: fill in the values and click “Save” at the bottom of the window — one button saves the whole window at once. You do not have to fill it in right now: “Set up later” keeps the skill installed, and you can configure it later from its card. However, “Cancel” and the × button, as long as nothing has been saved, also cancel the installation itself — the skill does not stay “half” connected. The secrets you configure belong to you and are not visible to other users.

If the settings show an error in the secrets declaration, the other fields remain available. Fields with empty or duplicate names cannot be filled in. Ask Iskra to fix the settings description of your skill and release a new version; for a catalog skill, contact its author, then update the installation. Keys and parameters that are already saved are not deleted when the error is shown.

If a skill requested setup in the middle of a task in chat, a “Set up” card stays next to the input field — it opens the same settings window for that skill and does not disappear after Iskra’s response is finished. Save the values, and Iskra will check

the connection and continue the interrupted task without you having to repeat the request. The card is kept for 30 days; the × button cancels the wait. Learn more in the “Agent mode” section.

The same skill can have different connections: a personal installation uses your personal key, and an organization installation uses the organization’s shared key. The shared key is set by the owner or an administrator; its value is not shown to members. If the skill is allowed only for selected groups, the shared key is available to the skill at run time only for members of those groups. Never send the key itself to the chat.

Exceptions are stated explicitly on the skill card. For example, with “Reputation.ru — personal key”, even an organization installation uses a separate key for each profile rather than the organization’s shared key. The installed personal-key variant replaces the system Reputation.ru skill and does not fall back to it if setup fails. Learn more in the “Reputation.ru: case law search” section. Wildberries Seller and Ozon Seller work the same way: an organization installation makes the skill available to members, but each profile connects its own seller account and stores its own credentials separately.

Consent to write

Some skills ask for the right to write values to their own secrets by themselves (for example, to save a token they have received). Before installing such a skill, Iskra shows a consent dialog with the list of slots — check each one and confirm. Without your consent, the skill cannot write to its secrets.

Overview

When a version newer than yours appears on the showcase, the card of the installed skill shows an “Update available” badge and an “Update” button — it moves your installation to the current version. Nothing is updated automatically.

If the new version changes the set of secrets the skill is allowed to write to, the card shows “Re-consent required” with a “Grant consent” button. Until you confirm, writing to secrets is blocked for the skill — this is a safeguard, not a malfunction.

Overview

On the “My skills” tab, in the “From catalog” section, open the skill card → at the bottom of the “Overview” tab, the “Skill management” section → “Delete skill”, and confirm the action. After that, the skill disappears from your catalog. Configured secrets are not erased: if you install the skill again later, they are picked up. The exception is external service accounts connected through the browser, such as Plaud: when you delete the skill from your personal skills, access to them is removed, and after reinstalling you need to connect the account again.

Overview

You can propose your own working skill (a personal skill or an organization skill — for an organization, the owner or an administrator can do this) to the shared catalog.

1. Open the “Skills” page and click your skill’s card. 2. At the bottom of the “Overview” tab, open “Skill management” → “Skill catalog” and

click “Publish to catalog”. This section is shown only when the catalog is enabled and your author or organization administrator permissions allow you to manage publication.

3. Choose a category, write a short description for the showcase card and

click “Submit request”.

Every publication — and every subsequent update — is reviewed manually by a platform administrator. While the request is being reviewed, the section shows the status “Request under review” and a “Cancel request” button. If the request is rejected, you will see the status “Request rejected” and the reason.

On the “Skills” page and in the chat with Iskra, a completed request with the status “rejected”, “build failed” or “cancelled” can be removed from the current profile’s list. This does not delete the request or the moderation history and does not hide them from other users who have access to them.

After approval, the section shows “In catalog: v…”. The showcase gets exactly the version the skill had when the request was submitted — your later edits do not affect it.

Other useful things to know:

Publication is an independent copy. Your skill remains yours: you can change it and even delete it — nothing breaks for those who installed the skill from the showcase.

Materials and connections. Shared files of the skill bundle are included in the published version. The author’s Memory folder bindings and connection settings are not transferred. Users attach Memory folders in their current profile and set up connections within their own permissions. The skills guide will help you choose where to keep templates, a price list and materials for a single task.

Updating a publication. Keep developing the skill as usual — changes do not reach the showcase on their own. When your version becomes newer than the published one (“In catalog: v1.2 · Yours: v1.5”), click “Submit update” — this is a new request that goes through the same moderation. The version must be newer than the published one; those who installed the skill will see the “Update available” badge and move to it whenever they want.

Limit: no more than 20 published skills per author.

Background skills (those that run on a schedule) are not yet accepted into the catalog.

You can remove a skill from the showcase yourself. While the skill is published, the “Skill catalog” section has a “Remove from showcase” button — after you confirm, the skill disappears from the catalog but keeps working for everyone who installed it. You can return it to the showcase with a new publication request (for the same version; moderation is carried out again).

If the skill has been removed from the showcase, the “Skill catalog” section shows the status “Removed from showcase” (you can submit a new request) or “Blocked by moderation”.

Overview

If you work in an organization profile with the owner or administrator role and the skill is not yet installed either personally or for the organization, after you click “Install”, Iskra asks where to install it: “Only for me” or “For your organization”. Before installing for the organization, choose the groups that will have access to the skill; an empty selection means the whole organization. The restriction is kept until the shared secret is configured, so the key is not exposed to members during installation. If a personal installation already exists, the button is labeled “Install for organization”; if the skill is already installed for the organization, the button remains “Install” and creates a personal installation. You cannot install a skill for the organization from a personal profile — switch to the organization profile. A skill installed for the organization appears for allowed members in the “Organization” section of the skill list in “Auto” mode; each member can choose “On command” or “Off” for themselves.

Organization installations are managed in the same place, in the “Organization” section of “Skills” (on the organization page, the “Skills” tab opens /skills directly): clicking a card opens a page that shows the version, the “Update available” badge with the

“Update” button (it moves the whole organization to the new version at once) and the delete button — after deletion, the skill disappears for all members.

If the same skill is installed both for you personally and for the organization, your personal installation takes effect — including its version.

Who sees the skill: groups

By default, an organization skill is visible to all its members. If you need to limit the audience — for example, give the skill only to the accounting team — use groups:

1. On the organization page, find the “Availability” line on the skill card: “Entire

organization” or “groups: N”.

2. Click “Configure” — the “Skill availability” dialog opens with the list of the organization’s

groups.

3. Check the groups that should have access to the skill, and save. An empty selection

means “Entire organization”.

This is how you configure the availability of both the organization’s own skills and catalog installations. If a selected group has been deleted, it is shown with the “(deleted group)” label — its former members do not see the skill; when all selected groups are deleted, nobody sees the skill, so remove dead groups from the selection.

A separate block, “Skills hidden from you by group access”, shows an administrator the skills whose group access hides them from the administrator’s own catalog — from there, the restriction can be changed or removed.

Overview

Iskra’s official Google Drive skill is in the “Integrations” category if the platform operator has configured and enabled the Google connection. The skill can search for files and view folder contents, read text files and Google documents, download a file into the conversation and upload a conversation file to Drive, create folders and text files, rename and move files, move them to the trash and grant access to a specific user. It also creates Google Docs documents and changes selected text in existing documents, reads, creates and edits Google Sheets, and builds and edits Google Slides presentations.

After installing it, ask Iskra, for example: “find the June report on Google Drive”. On the first call, Iskra offers to connect your Google account. Until the Iskra app has passed Google’s verification, a separate warning appears first, and then Google may show the “Google hasn’t verified this app” screen. Continue only if you trust your Iskra instance and agree to give it access to your files. In Google’s testing mode, only accounts that the operator has added to the test users list can connect.

Do not send a token or password as a chat message. If browser-based connection is unavailable for an already installed skill, the skill may offer a fallback secure form for an access token or OAuth data; use it only as instructed by the administrator of your instance. This form covers only work with Drive files: Google documents, spreadsheets and presentations do not work through it — if browser-based connection is not configured on your instance, Iskra will say so and suggest contacting the administrator. A connected Google Drive is enough for them, though: Iskra offers a separate connection to documents only if nothing is connected.

Search and folder contents

You can search by part of a name, or ask Iskra to show what is in a specific folder: “show the files in the Clients folder” — and then narrow it down inside that folder: “find the contract there”. The

response shows the modification date, size and a link to each file, so you can tell a fresh document from last year’s right away, without asking about each one separately.

How to download a file into the conversation

Reading shows Iskra only the text, so it is better to download a PDF contract, a spreadsheet or an image right away: ask, for example, “download the company details sheet from Drive”. The file appears among the conversation files, and Iskra tells you its name and size. Then you can work with it like any file you brought yourself — for example, ask Iskra to recognize the document or attach it to a deal in amoCRM.

Google documents, spreadsheets, presentations and drawings are saved in familiar formats: Word, Excel, PowerPoint and PDF, respectively. If you need a different format, say so — for example, “save the spreadsheet as a PDF”. Other files are downloaded as is.

How to upload a file to Drive

The reverse works too: “save this contract to Drive in the Clients folder” — Iskra takes a conversation file, either yours or one it created itself, and puts it on Drive. PDFs, office documents, images and archives are supported; the limit here is lower than for downloading: 3 MB.

File size is limited to 20 MB. A larger file or an entire folder is easier to get in Google Drive itself — Iskra will tell you so rather than leave you without an answer.

Renaming happens immediately. Before moving a file to another folder, Iskra shows which file will be moved and where; for an old file that is in several folders at once, you will need to explicitly choose the previous folder. Moving to the trash and granting access to another user are also done only after your confirmation. Closing the form or a timeout counts as a refusal: no changes are made on Google Drive.

Creating and editing Google Docs

Ask, for example: “Create a Google document with a meeting plan, headings and a bulleted list” or “In this document, on the ‘Plan’ tab, replace the second paragraph and add a summary at the end”. Iskra can create empty documents, add paragraphs and change or delete a selected fragment. Headings and bulleted lists are supported; tables, images and other tabs outside the fragment being changed are preserved. Creating tables, images and tabs is not part of these operations. In one call you can pass up to 100 paragraphs and 32 KiB of text, and the title of a new document can be up to 200 characters. The full preview must also fit into the confirmation form. If the text contains many special characters or the titles are too long, Iskra asks you to reduce the size of the edit or the length of the titles. Until then nothing is written, and the preview content is not truncated. Targeted text replacement works within a single paragraph. To change several paragraphs, Iskra offers to replace the paragraphs entirely.

The first time you use these operations, Google separately asks for access to documents. Before each write, Iskra shows the prepared content; for an edit, it also shows the document, the tab and the selected range. The change is made after the owner of the connected profile clicks the button. This confirmation is required even in “Allow and don’t ask again” mode; another participant in a shared chat cannot confirm the write on the owner’s behalf. Without confirmation, no changes are made.

Before editing, Iskra checks the document version. If someone has changed it in the meantime, the document has to be read again and the new proposal reviewed. If the document was created but filling in its initial content failed, Iskra reports this and returns the address of the created document. If Google’s response is lost, the result may be unknown:

first check the document in Google Docs, then decide whether a new request is needed. A repeated write or the creation of another document is not started automatically.

Google Sheets

Ask, for example: “show what is in the estimate workbook”, “build a table of revenue by month” or “append these three rows to the Data sheet”. Iskra reads a workbook through a window: it shows the sheets, their sizes and the values of the selected range — for example, “Data!A1:D50” — and continues by itself from where it stopped if the range did not fit. It distinguishes “not everything is shown” from “the rest is empty” and says so directly.

A new workbook is assembled from sheets and rows, optionally right in the folder you need. In an existing workbook, one action is performed at a time: appending rows, inserting rows or adding a sheet happen without a separate question; writing values to a range, clearing a range, deleting rows or deleting a sheet happen only after your confirmation, because these actions erase what was there. Iskra preserves value types: text stays text, a number stays a number, and Sheets calculates formulas itself.

Google Slides

Ask Iskra to build a deck from an outline — “make a five-slide presentation on the quarter’s results” — or from your template: Iskra copies it and substitutes values for the tags, keeping the design. In a finished deck, it adds and reorders slides, replaces the text of a selected box, table cells and speaker notes, and deletes slides. Replacing text, replacing notes and deleting slides are done after your confirmation. If someone changes the presentation while you are reviewing the proposal, Iskra tells you so and offers to re-read the deck rather than overwrite someone else’s edit.

What the confirmation shows

Before writing to a spreadsheet or presentation, Iskra shows a form with a fragment of the file’s content: addresses, cell values, rows of the sheet being deleted or slide text. This form is posted to the current conversation, so in a shared chat other participants will see it, and it stays in the conversation history and export even after access to the file in Google is closed. Only a trace of the edit is shown: long cell values are shortened, and for a sheet being deleted, the first few rows are visible.

Only the person who invoked the tool, and only the owner of the connected Google account, can click the button: another participant in a shared chat cannot confirm the write on your behalf, and this confirmation also remains in “Allow and don’t ask again” mode.

Sheets and presentations work over the same Google Drive connection — no separate consent is needed for them. If the administrator of your instance has not yet enabled access to Google Sheets or Google Slides, Iskra says so directly and suggests contacting them: your account’s access to the file has nothing to do with it.

Overview

Iskra’s official MAX skill is in the “Integrations” category if the platform operator has enabled it. It connects not a personal account but a MAX chatbot that you have created. No separate launch or scenario is needed: install the skill and write in a regular chat, for example, “Help me create a MAX bot” or “Connect my MAX bot”. Iskra finds out what stage you are at, explains the next step and calls the skill’s tools as needed. An invite link to a chat does not connect anything on its own: you need a connected bot that has been added to the chat with the required permissions.

Creating and moderating a bot

Iskra explains the steps but does not create the profile or the bot and does not influence the moderation decision. You perform these steps on the MAX for Partners platform.

According to the official MAX documentation, checked on August 3, 2026:

a bot can be created by an organization, a sole proprietor or a self-employed person who is a resident of the Russian Federation and has a verified profile;

an organization or a sole proprietor can create up to 5 bots, a self-employed person up to 2; the creation path is: the relevant profile → “Chatbots” → “Create”;

a logo, name and description are required for the card; Iskra suggests checking the current sizes, formats and restrictions before submission;

every bot goes through moderation; the stated time is up to 48 hours on business days;

the status “under moderation” means waiting, “created” means the review was successful, and “requires corrections” means you need to open the comment on the card, fix the data and submit it again.

If the bot limit has been reached or the profile does not meet the requirements, Iskra says so directly and does not suggest a workaround. If the review takes longer than the stated time, check the notifications and the status on the platform and contact the official MAX support. Changing the data of an already created bot triggers moderation again.

Current requirements and rules:

creating and moderating a bot;

managing a bot and getting a token;

chatbot FAQ;

placement rules.

Token and connection to Iskra

After the bot passes moderation, open “Chatbots” → “Go to” → “Advanced settings” → “Configure”. This is where you find the token and a toggle that allows adding the bot to group chats; such adding is disabled by default.

The token’s storage scope is chosen when you install the skill: personal or shared across the organization. Only the owner or an administrator can connect and disconnect the shared organization connection; Iskra tells other members this directly and offers to install the skill personally if they need the bot just for themselves. In the catalog, a skill already installed by the organization is marked “Installed in organization”; the install button next to it installs the skill for you personally.

Do not paste the token into a regular message to Iskra. Ask Iskra to connect MAX: if the token is not set yet, Iskra shows a “The skill needs secrets” card with a “Configure” button below its reply. The card is visible even when the timeline is collapsed; in the timeline itself, the same form opens with the “Configure skill secrets” button with a lock icon. Clicking it opens the secure form of the installed skill. With a personal installation, the token is saved for your profile; with an organization installation, it goes into the organization’s shared form. If the token does end up in the chat, it stays in the chat history, so consider it compromised: regenerate the token in the MAX advanced settings and save the new one through the secure form. If the new token belongs to a different bot, the next time the skill is used it clears the previous bot’s connection and chat list and then connects the new bot. Data of the two bots is never mixed.

When you set up the skill for the first time, once you successfully save the form opened from the chat, the original chat automatically resumes the previous task: Iskra connects or checks the bot as you requested. A planned replacement of an already saved token sends nothing to the chat, and neither does saving the token on a separate

settings page; after replacing it, ask Iskra to connect or check MAX. You can then add the bot to the group or channel you need. A private chat with the bot is connected separately for each Iskra profile; this keeps organization members from accessing each other’s conversations.

Once connected, Iskra works only with the group chats and channels the bot has been added to and where its permissions allow the required action. With an organization installation, these groups and channels are visible to all organization members who have access to the skill. Private chats with the bot are not shown by default. The bot owner’s personal account and the owner’s conversations with other users or bots are not available to the skill. If the bot is not a member or an administrator of the chat you need, the skill does not bypass this restriction.

Private chat with the bot

To work with your private chat from Iskra:

1. Open the private chat with the connected bot in MAX.

2. Write to Iskra: “Connect my private MAX chat”.

3. Iskra shows a secure form with a one-time code. Send this code

as a separate message to the private chat with exactly the specified MAX bot.

4. Return to Iskra and click “Verify link”. The code is valid for 5 minutes.

You don’t need to enter your name or numeric MAX ID: only the message with the one-time code proves that you own the chat. After linking, the current Iskra profile sees the history and can write to this private chat on behalf of the bot. Other organization members won’t see it. If you link the same chat to another Iskra profile, the previous link stops working. The command “Disconnect my private MAX chat” removes only your link and does not disconnect the shared bot, groups or channels.

To connect a group or channel:

1. In the bot’s advanced settings, allow adding it to group chats.

2. In MAX, add the bot to the chat you need as a member; you can find it by its nickname or

the public link from the bot card.

3. Make the bot an administrator and grant it the read_all_messages permission. Without it, the bot does not

receive new group or channel messages, so the chat may not appear in Iskra, and the available history will be incomplete.

4. Grant only the permissions your tasks need: write — editing and deleting

messages in a group and posting in a channel; edit and delete — editing and deleting channel posts; pin_message — pinning; change_chat_info — changing the chat; add_remove_members and add_admins — managing members and administrators.

MAX does not provide a full list of all the bot’s chats, so Iskra’s list fills up as new activity happens. After adding the bot, send any message to the group or channel, then ask Iskra to show the available chats. If you know the chat_id, ask Iskra to add the group or channel by it; Iskra first safely checks that the chat is accessible to the bot. Groups and channels may have a negative ID. Iskra stores up to 500 of the most recent known groups and channels; if there are more, the overall list may be incomplete, but you can check and add the chat you need by its chat_id. The list shows only chats where the bot is currently active. Reading messages and performing actions in a known chat go directly through MAX.

The skill lets you read known chats and messages; send, edit and delete messages; and manage members, the chat name and the pinned message within the limits of the MAX Bot API. When reading history, 50 messages is the default, not the maximum: one request can return up to 100 messages. For deeper history, Iskra can make further requests with time bounds. The skill shows the available history that MAX returns for this bot and chat;

it does not necessarily start from the moment Iskra was connected. Before changing a message, chat, member or pin, Iskra shows a separate confirmation with the exact target. If the response is uncertain, the change is not retried automatically. Before deleting, the skill also checks the recipient of the message itself: chat_id for a group or channel and recipient.user_id for a private chat with the bot.

Before uninstalling the skill, it’s best to ask Iskra to disconnect MAX: the skill deletes the connection data. If the previous bot’s token is no longer available, revoke it in the MAX advanced settings if needed.

MAX files and attachments

If a MAX message contains a file or image, it appears in the message list with its number, name and size. Ask Iskra to fetch the attachment, and the file appears in the current chat’s files. Links and previews don’t count as attachments: they have no file and don’t appear in the list.

You can send a file too: attach it to your message to Iskra (or use a file that is already in this conversation) and ask Iskra to send it; it goes out as an attachment with its name and type preserved. This means the files of your conversation with Iskra, not files in the MAX chat: you don’t need to upload anything to MAX beforehand. Before sending, Iskra shows a confirmation with the exact chat, file name and file size; the message goes out only after the file has fully uploaded, so an unfinished upload never turns into a sent message.

The file limit is 3.5 MB in both directions; a larger file is rejected upfront with a clear error.

Overview

Iskra’s official Telegram skill is in the “Integrations” category if the platform operator has enabled it. It connects not your personal account but a chatbot you have created. No separate launch or scenario is needed: install the skill and write in a regular chat, for example “Help me create a Telegram bot” or “Connect my bot”. Iskra figures out which stage you are at, explains the next step and calls the skill’s tools as needed.

Creating a bot and getting the token

You create the bot yourself with @BotFather; Iskra explains the steps but does not do this for you. According to the official Telegram documentation, as checked on July 30, 2026:

open @BotFather and send the /newbot command;

set the display name, then the username, which must end in “bot”;

in reply, you receive a token in the form <digits>:<letters and digits>.

If the token is lost, reissue it with the /token or /revoke command: both issue a new token and invalidate the old one, and there is no way to view the old one. After either command, save the new value through the secure form; otherwise the connection stops working.

Do not paste the token into a regular message to Iskra. Ask Iskra to connect Telegram: if the token is not set yet, Iskra shows the “Configure skill secrets” button with a lock icon, and clicking it opens the secure form. The token’s storage scope is chosen when you install the skill: personal or shared across the organization. If the token does end up in the chat, consider it compromised: run /revoke with @BotFather and save the new token through the form. If the new token belongs to a different bot, the skill clears the previous bot’s conversation registry, saved messages and event queue and asks you to reconnect; data of the two bots is never mixed.

How Iskra receives events

When connecting, Iskra gives the bot its own secure address and asks Telegram to send events there. From then on, events arrive as soon as something happens and accumulate in your installation’s queue. The skill processes the queue at the moment you turn to it. Your presence is needed for processing, but not for receiving.

This has an important consequence: a bot can have only one such address. If your bot is already set up to deliver events to another service (your own server, n8n and the like), connecting it to Iskra replaces that setup. This never happens silently: Iskra shows the current address and warns that the integration using it will stop receiving events. Nothing changes without your consent. The reverse is also true: if, after connecting to Iskra, the bot is pointed at another service, events stop arriving, and the bot status check will report it.

Unprocessed events are kept for a week. If the installation queue overflows, Telegram temporarily cannot deliver new ones: it keeps retrying, and whatever is not accepted is lost after a day. The bot status check shows how many events Telegram failed to deliver and reports delivery errors.

Groups, channels and privacy mode

To work with a group or channel, add the bot to it. An invite link to a group or channel does not by itself connect the chat to Iskra: the bot must become a member of the chat and get the required permissions. By default, a bot in a group sees only commands and messages addressed to it. For the bot to read all messages:

1. run /setprivacy with @BotFather and choose Disable, or make the bot

an administrator of the group;

2. after changing the privacy mode, remove the bot from the group and add it again;

otherwise the setting won’t take effect.

You can check the current state by asking Iskra to show the bot status: the reply indicates whether the bot reads all group messages.

One bot, one installation

The token’s storage scope is chosen when you install the skill: personal or shared across the organization. Only the owner or an administrator can connect and disconnect the shared organization connection. If you don’t have these rights, Iskra tells you so and offers to install the skill personally: the bot will then be yours rather than shared by the organization. In the catalog, a skill already installed by the organization is marked “Installed in organization”; the install button next to it installs the skill for you personally.

What you must not do is connect the same bot to two installations. A bot has only one event delivery address, so the second installation intercepts the first one’s events, and the first one goes silent. Different installations need different bots.

What organization members see

Groups and channels are shared. As soon as an administrator adds the bot to a chat, that chat appears for everyone who has access to the installation. Nobody needs to reconnect, and nobody enters their own token.

Accumulated group messages, however, are collected separately by each member: they start accumulating from the moment that member first turns to the skill. The Bot API has no full history at all, so the message window is declared incomplete anyway.

Private chats are not shared. The bot talks to many people, and each of its private conversations belongs to the person in it. By default, such a chat is unavailable to anyone, including the person who connected the bot: it doesn’t appear in the shared list at all.

Telegram Business conversations are not shared either. Business account chats are visible only to the member who has confirmed ownership of that Telegram account with a one-time code. Others can’t see them and can’t reply in them on someone else’s behalf, even if the bot is shared.

How to connect your private chat with the bot

Ask Iskra to connect your private chat. It shows a secure form with

1. open the link, and Telegram itself sends the code to the bot;

2. if the link doesn’t open, send the same code as a regular message to the private chat

with the bot;

3. return to the form and confirm.

The code is valid for 5 minutes and must not be sent anywhere except Telegram: for those five minutes, it is equivalent to access to the chat. Don’t forward it to anyone and don’t paste it into an Iskra chat.

After that, the chat appears in the list and works like a regular conversation. It stays visible only to you: your colleagues in the organization won’t see it, even if they have access to the same bot.

If you need to undo this, ask Iskra to unlink the chat. The accumulated message window is then deleted, while the conversation itself remains in Telegram. If another person links the same chat again, the previous link is canceled: a chat belongs to one owner.

Secretary mode: Telegram Business

The bot can work with the private chats of your business account. You set this up on your side. Telegram Business features (opening hours, quick replies, greetings, business links) are part of the Telegram Premium subscription, but connecting a chatbot is exempt from it and works without a subscription. Only one bot can be connected to an account; the previous one is replaced. If there is no “Telegram Business” section in the settings, update the Telegram app. Steps:

1. in @BotFather: /mybots → your bot → Bot Settings → Secretary Mode (formerly

called Business Mode) → turn it on;

2. in Telegram: Settings → Telegram Business → Chatbots → enter the bot’s username and

grant permissions.

Business gives access to the private chats of the business account, but not to groups and channels: you still add the bot to those manually. Business conversations are visible only to you. For Iskra to tell the account owner apart from other organization members, link your Telegram to your profile with a one-time code, the same way as for a private chat with the bot. There is a single shared link: if you have already connected a private chat, you don’t need to go through the code again, and a link made for Business also opens your private chat to you. The only requirement is that Business is connected from the same Telegram account that is linked to your profile. Until the link exists, chats from this connection are closed to both reading and replying. This restriction doesn’t get in the way of a team: one bot can have many connections, and each member connects and links their own account. Connect the bot in the Telegram Business settings and then turn to the skill: the notification arrives at the moment of the change and is not repeated on its own. If Iskra says it doesn’t see any business connections, connect the bot in the settings again. The bot can reply on the owner’s behalf only in chats where the other person has written within the last 24 hours, and only within the granted permissions: the window is opened specifically by an incoming message, and your own reply does not extend it. If the reply permission isn’t granted or the connection is turned off, Iskra says so directly and won’t try to bypass the restriction.

What the skill can and cannot do

The skill reads known chats and accumulated messages; sends, edits, deletes and pins messages; shows administrators and the member count; bans and unbans members; and changes the chat name. Before any change, Iskra shows a separate confirmation with the exact target and the full text. If the response is uncertain, the operation is not retried automatically.

The skill also works with files. If a message contains a document or image, it appears in the message list with its name, type and size. Ask Iskra to fetch the attachment, and the file appears in the current conversation’s files, where you can analyze it like any other document. In the other direction: attach a file to your message to Iskra (or use a file already in the conversation) and ask Iskra to send it; it is sent to the chat as an attachment with its name and type preserved. You don’t need to upload the file to Telegram beforehand. Before sending, Iskra shows a confirmation with the destination, file name and file size.

The file limit is 3.5 MB in both directions. This is Iskra’s limit, not Telegram’s: a larger file is rejected right away with a clear error, before downloading or before the send confirmation, and the limit cannot be bypassed.

The main limitation is that the Bot API has no message history. Iskra sees only messages received after the connection and keeps a limited window of recent messages for each conversation. An empty message list means no events have arrived, not that the chat is empty. The chat list is also built from events: the Bot API doesn’t provide a list of the bot’s chats, so an existing chat appears after a new event or after you explicitly add its chat_id. The registry is limited to 500 chats and a total record size, so with long chat names the actual limit is lower; on overflow, the skill openly reports the truncation.

Events don’t run Iskra in the background: they arrive immediately but are processed the next time you turn to the skill.

Before uninstalling the skill, ask Iskra to disconnect Telegram: the skill then removes the event delivery address from the bot and deletes local data, namely the conversation registry, saved messages and event queue. The bot itself stays in @BotFather, and you can connect it again or connect it to another service. If you skip this step, Telegram keeps sending events to an address that no longer exists.

In an organization installation, saved messages are stored separately for each member, and disconnecting erases them only for the person who did it. The others just need to ask Iskra to disconnect Telegram for themselves: the skill sees that the bot is already disconnected and deletes their copy. If the bot is connected again, the copy is deleted automatically the first time the skill is used. It can’t be read before then anyway.

Overview

The first-party skill “AI media tools (your key)” is in the “Productivity” category. It is the same image, video and voice-over generation skill that everyone has, but it uses your personal OpenRouter key: no Iskra megatokens (MT) are deducted for generation; the provider charges your account. The key is requested at installation; if you don’t set it, the skill itself opens a secure form in the chat. Requests with your key go to the provider address set by the installation administrator: it may be an OpenRouter mirror or a corporate proxy rather than openrouter.ai directly; a direct connection is not guaranteed. The administrator can see the address in the skill settings. While the BYOK skill is installed, the regular system version is not shown in your skill list; to go back to it, delete the BYOK skill: there is no automatic switch back, including when the key stops working.

Capabilities and limits match the system media skill: up to four separate images per request, up to five input images in total for composition or

editing, video from the first/last frame or from references, and voice-over with delivery instructions and from a voice sample. Input files from the current chat are preserved. Recording transcription is performed by Iskra’s shared tool if it is available in the chat.

Voice-over uses the same settings of the selected model: Gemini and Fish accept delivery instructions, while MAI-Voice-2 accepts a speed from 0.5 to 2. Gemini saves WAV and accepts up to 1,000 characters per call; the other models save MP3 and accept up to 4,000. The catalog shows the available settings, and an unsupported instruction requires choosing a compatible model.

With a partial result, the finished files remain available; successful variants are not retried automatically. Stopping a request does not guarantee that generation stops at the provider: it may have charged you even for a variant with an unknown outcome. Before retrying, check the finished files and your spending in your personal OpenRouter account.

Overview

The first-party “Music notation” skill is in the “Productivity” category. Once it is installed, Iskra creates and edits sheet-music scores in MusicXML/MXL format through a structured model and checks the file before delivering it: validation against the official W3C MusicXML schema; re-reading the file and comparing the note events, durations, and each part’s initial time signature, key and clef with the model prepared before the file was written; measure consistency; and a graphical render of the first page to SVG, which you can open right in the chat. In its report, Iskra names exactly what was checked; a match with the prepared model does not prove that the original transcription is accurate or that all data survived import. On request, the skill also exports MIDI (checking re-reading and the presence of notes) or MEI (re-reading and a Verovio render) along with the verified MusicXML/MXL; preservation of all properties in these formats is not guaranteed. Accurate transcription of notes from audio, an image, a PDF or video is not available with or without the skill: for such a request, Iskra states the limitation directly and offers a choice between declining and an approximation explicitly labeled as such. Before delivering MusicXML/MXL, the skill checks that this label was preserved in the header on export; other metadata requires a separate check. Byte-for-byte copying of the source does not by itself mean the score has been verified.

Overview

Iskra’s official UniSender skill is in the “Integrations” category if the platform operator has enabled it. To connect, you need a UniSender API key: with a personal installation, set your own key; with an organization installation, the owner or an administrator sets a shared key. If there is no key yet, Iskra itself opens the skill settings form in the chat, so you don’t need to look for it in the menu. You can build and preview an email without a key: it is needed only where the skill calls UniSender.

The skill can check the connection and answer almost any question about your UniSender data: lists, tags, personalization fields, templates, campaigns and their metrics, created emails, sender domains. It shows how many contacts are in the database in total, how many are active and how many addresses can be mailed at all. The UniSender API does not expose plan limits, so you’ll have to compare these figures with your plan yourself. In this skill, the word “segment” means a UniSender list; a tag remains a separate optional filter. If a name could refer to either a list or a tag, Iskra should ask which one you mean. A large export is processed inside the skill: the temporary link and the full contact file are not passed to the chat.

The skill itself builds the email, not Iskra: you describe what the email should contain and get ready-made HTML markup. For an email based on your UniSender template, the skill changes only the texts and link addresses and leaves the markup itself untouched. A finished layout from a designer works too: attach the .html file to the chat, and the skill takes it as is;

it warns about defects it notices in third-party markup but won’t stop you from sending. Separately, the skill warns about UniSender tracking links (geteml.com/…/mail_link_tracker) that end up in a template from an already sent email: UniSender won’t accept such an email, so the skill decodes the original address and suggests to Iskra what to replace the link with. If UniSender still rejects the email, a test send or another change, Iskra sees the reason in the provider’s own words, not just an error code. The email code is not printed in the chat. The finished email appears as the email-preview.html file in the chat files side panel, where you can open and view it; no separate confirmation window is needed for this. External images in the preview are not loaded by default, so that the email’s sender doesn’t learn that you viewed it and doesn’t get your IP address: you can show them with the “Show images” button. The file preview and the full-screen view of an HTML artifact behave the same way; the button is available on all three screens. Images you attached are visible right away. Do not open HTML through a manually composed link to an internal path. For a test or bulk send, the separate confirmation window remains and shows the email actually being sent, the recipients or lists, the links and the images. What you saw and confirmed is exactly what goes to UniSender.

Templates and senders

You can save a built email back to UniSender as a template; it then opens in the UniSender editor and can be reused next time. A new template is created without extra questions; overwriting an existing one erases the previous markup and therefore goes through the confirmation window.

If the sender address you need is not among the confirmed ones, the skill asks UniSender to send an email with a confirmation link to that address, so you no longer need to go to your UniSender account for this. Until you follow the link, sending from this address will be rejected.

Editing an already built email doesn’t require rebuilding it entirely: Iskra changes only the named block or appends new ones at the end. You don’t need to attach the images again for this, as they are stored with the email; attach a new image only if you are adding a new block with a picture.

Lists, contacts, fields and tags

The skill doesn’t just read UniSender reference data; it also changes it. Lists can be created, renamed and deleted. Contacts can be added to lists, have their field values and tags updated, be removed from individual lists or be unsubscribed from all mailings at once. Personalization fields and tags can be created, renamed and deleted.

Tags and field values can be set for all contacts in a call at once. A field value set for an individual contact overrides the common one, while tags are combined. An update changes only the listed fields; the contact’s other fields and lists stay as they were. Tags, however, are replaced entirely on update: the contact keeps exactly the tags you named. So to remove one tag, list the ones that should remain, and to remove all tags, say so: the skill will send an empty set. If your request doesn’t mention tags at all, they stay as they were. You can see the current tags in the contact list.

Iskra takes the addresses for these operations from the conversation: you can name them yourself or first ask Iskra to find the contacts you need and then say “add them to the list”. In one go, the skill removes from lists and unsubscribes no more than 50 addresses, and adds and updates up to 500. Above 50 addresses, adding and updating go through UniSender’s bulk import. Bulk import can only replace field values, so if contacts have different sets of fields, the skill asks you to split the call so as not to erase data the request didn’t concern. Bulk import can only add tags to the existing ones: you can tag hundreds of addresses by adding, but replacing a tag set is done one contact at a time and is therefore limited to fifty addresses at a time. Uploading an entire database is still done by import in your UniSender account.

If UniSender doesn’t respond to a change (creating a list, renaming, deleting, overwriting a template or uploading contacts), the skill doesn’t retry the call blindly and doesn’t give the usual “try again later”. It says directly that it is unknown whether the change was applied and asks you to check the state first: it won’t guess on its own, because a list with the same name may have existed before. Otherwise the database would end up with two lists with the same name, and sending by such a name would no longer be possible. For bulk additions, the skill reports the number that UniSender itself counted: if UniSender didn’t accept all addresses, the skill says so directly.

The confirmation window appears where the consequences can’t be undone or many contacts are affected: deleting a list, field or tag, unsubscribing, removing from a list, and editing ten or more contacts at once. Replacing tags is always confirmed, even for a single contact: the skill doesn’t store the previous tag set and can’t restore it. The window shows not only the addresses and lists but also exactly what is being applied: which fields with which values and which tags; if contacts have different values, it says so. This matters on update: the previous values of the specified fields are replaced, and you won’t have to confirm the replacement blindly. Creating a list, renaming and editing a couple of contacts go through without the window. If some addresses couldn’t be changed, the skill honestly names both the processed and the failed ones. It separately names the addresses for which UniSender didn’t respond: the change may have been applied there, so don’t retry blindly; check their state first.

Separately, the skill can build a mailing list when there are more addresses than you can list by hand: merge several lists, subtract one or two, select by tag or status, or take those who reacted in a certain way to a past mailing: opened it, clicked a link, opened but didn’t click, received but didn’t open. The addresses themselves are not exposed: Iskra names only numbers, and the selection itself becomes the content of the chosen list. The skill removes duplicates itself, and you can build one list from several sources in a row without duplicates appearing. If no status is set, the list gets only addresses that UniSender will mail to. You can first ask just to count how many there would be, without writing anything. You can name up to twenty source lists and subtract as many, or take the whole database, in which case at least one other filter is required; the target list cannot be a source. The default and maximum limit is 20,000 addresses, and it applies to the resulting size of the list, not to a single selection: if the list already contains addresses, the skill adds only up to the remainder; this is what directly addresses the plan’s monthly limit, which is calculated by list size. Addresses already in this list are not written again, and Iskra tells you how many there were, but only when the source is lists, a tag or a status. An export by reaction to a mailing doesn’t contain list membership, so in that case the skill honestly warns that some of the selected addresses may already have been in the list and won’t give an exact number. Before writing, a confirmation window appears with the source, the number of addresses and the current list size. If UniSender didn’t report how many contacts are in the list, the skill warns directly that it can’t guarantee the limit this time. And if, after writing, the provider reports that the account now has more contacts than the plan allows, you will be told about it rather than discovering it at the next mailing. UniSender keeps the delivery report for 180 days, so a reaction-based selection can’t be built for an older mailing.

There is no reverse operation: you can build a list by criteria, but you can’t remove addresses by criteria. Removal from a list is done one address at a time, no more than fifty per call, because UniSender doesn’t offer bulk removal from a list. Along the same lines: the skill doesn’t run A/B tests or automated email sequences and doesn’t wake up on a schedule by itself; a mailing goes out when you scheduled it.

A new contact is added without a confirmation email and gets the “new” status in UniSender. It never becomes confirmed through the API, whatever the settings: only the contact can do that with their own action. Responsibility for having consent to mailings remains with the database owner, and you can always see the actual statuses by asking Iskra to show the list’s contacts.

After a test send, the skill confirms only that UniSender accepted the request. The Legacy API doesn’t report the delivery status of a test email, so check the result in the mailbox; the campaign status doesn’t apply to a test send.

Mailing history is available separately: what was sent and when, to which lists and with which statuses. One email sent to several lists is shown as one mailing rather than as three entries. Metrics (deliveries, opens, clicks, unsubscribes, complaints) are calculated for the mailing as a whole; for a mailing in progress, the numbers are marked as incomplete rather than hidden.

The skill can also fetch up to 500 sender domains by login and show up to 20 name and email pairs from recent campaigns. This history is not a complete list of currently confirmed addresses, but the skill can read the complete list separately, and it finds out the account login by itself. Iskra should not make up address variants.

A mailing goes through a safeguarded scenario:

1. Choose one to three lists and an email: a ready-made template or new text.

The skill rejects empty links, unknown link templates and unsafe addresses. The confirmation shows exactly the links and external images from the final email.

2. You can add PNG, JPEG or GIF images to the email, up to thirty-two of them, each no

larger than 300 KiB: attach the file to the current chat. A separate overall limit applies to the full encoded request, so several large images may need to be downscaled or compressed. Iskra shows the images before sending. The skill doesn’t search for or download images on the internet, because it can’t verify the license and usage rights. External images already in the selected template are kept.

3. Iskra shows the sender, subject, list names, actual links and

images, as well as a brief audience estimate. Choose “Send now” or “Schedule”; the mode and time fields are located above the visual HTML preview. If one address is in several lists, it may receive the email several times. A list with an estimate of more than 50,000 recipients must be sent separately; the form warns that processing may take longer. Sending immediately requires one more explicit answer: “Yes, send now”.

4. The close button (×) or Escape only collapses the full-screen window: click “Return to

form” in the tool card to open the same preview. The “Cancel” button ends the call without creating any emails or campaigns.

5. After the final confirmation, the skill first creates a separate email for

each list, then separate campaigns. For “now”, UniSender launches them without a delayed start time; when scheduling, all campaigns get the selected time.

To check the email, you can separately send a test to 1–25 explicitly specified addresses. An existing UniSender list is needed only as a technical requirement for creating the email: addresses aren’t imported into it, and no campaign is launched for the list. Iskra shows a single confirmation with the exact email, recipients, links and images. UniSender may reject an unconfirmed test address. If the response is lost or ambiguous, the skill honestly reports that the result is unknown and won’t repeat the send automatically.

The result contains a short text message with the sender, subject, actual links and known campaign IDs, so this information stays in the chat context without raw HTML being printed. When asked “What is the status of the email?”, the agent re-reads the current state of the known campaigns in UniSender. You can cancel a scheduled mailing by its name, its place in the context or the campaign ID: the agent shows a separate confirmation and cancels all related campaigns that can still be canceled. If the result is unknown, it checks the fresh status and won’t blindly retry the cancellation. UniSender doesn’t allow canceling a send that has already started or finished.

When asked to analyze recent mailings or up to ten specific campaigns, the skill compares aggregated UniSender metrics: deliveries, opens, clicks, unsubscribes and spam complaints. To explain the result, it can use the subject, a short excerpt from the email’s plain-text version and popular links that have been safely sanitized. This does not include reading recipients’ free-form replies: the UniSender Legacy API doesn’t provide them. Recipients’ addresses and IPs are not included in the analysis result; if some of the content or links are unavailable, Iskra should state this explicitly and keep the metrics already obtained.

accepted=true means that UniSender definitely accepted all targets;

“nothing was created”. The authoritative field for each target is state: accepted, failed, unknown or not_attempted. With a partial or unknown result, don’t retry the whole original set of lists: first ask the agent to check the returned campaign_id values and, if needed, verify them in your UniSender account, then start a new call only for the lists that were not sent.

On UniSender’s “Free” plan, the service itself adds a “Made in UniSender” logo at the bottom of every email; it is not part of the HTML built by Iskra. On paid UniSender plans, this block is not added; see the UniSender plan descriptions for details.

Overview

“Contacts” connects your external CardDAV address book. Install the skill from the catalog and open its settings: enter your login, app password and the HTTPS server address. If the provider supports auto-discovery, the full login with the domain and the password are enough. Tested connections are Yandex and Radicale 3.5.10, with basic authentication by login and app password. Compatibility with other servers requires separate testing.

The connection is set up separately from mail, calendar and the WebDAV drive. You can use personal settings or the organization connection. Before writing, the form shows the account and the selected address book.

Ask Iskra to find a person by name, organization, phone or email. A contact card includes the name, organization, job title, birthday, a note, and several phone numbers and email addresses. If there are many results, Iskra shows a short list; you can open the selected card in full. If Iskra reports that the search is incomplete, narrow down the query or specify the address book. A card that is unsupported, unavailable or deleted during the search is skipped; the other matches remain visible, and the search is marked as incomplete. An address book access error means you need to check the connection. If the server has a very large number of address books, Iskra shows part of the list and warns that it is incomplete; in that case, it’s easier to name the book you need.

To create, edit or delete a contact, check the data in the form and confirm the action. If several address books are available, choose the one you need first. Canceling the form leaves the address book unchanged. If another client has changed the card in the meantime, Iskra offers to open its fresh version and confirm the edit again.

For a large card, long fields in the form may be shown abbreviated; a warning above the preview states this explicitly. The confirmation applies to the full contact data; the abbreviation affects only the display.

A contact without a display name is available just like the others: some services store such records, and Iskra shows them by organization, phone or email address. You can’t set an empty name when saving.

The skill preserves the card’s other fields when making a targeted change. vCard 3.0/4.0 cards in UTF-8 are supported. A card with an unsupported field can’t be read or changed and is skipped in search with a notice that the result is incomplete. This applies to vCard 2.1, other encodings, and an unsupported parameter or value type

in the name, organization, job title, birthday, note, phone or email. For example, an old phone entry TEL;WORK;VOICE instead of the TYPE=work,voice parameter, or an email with the VALUE=uri type, makes the whole card unavailable. Re-save such a contact in a modern format through your contacts app. Contact groups, bulk import and continuous sync are not part of this skill.

Overview

The Plaud skill works with recordings from the Plaud voice recorder: it finds recordings, reads the meeting summary and transcript, helps you break down the meeting and saves recordings to Memory. This is not a system skill: it doesn’t appear for you automatically and can be installed only from the skill catalog, in the “Integrations” category. The skill is in the catalog only if your Iskra administrator has enabled the Plaud connection. If the administrator later disables the connection, the skill disappears from the catalog and the chat, but the installation is kept until the connection is enabled again. If Plaud doesn’t accept Iskra’s connection settings, the skill asks you to contact the administrator.

From the catalog, you install the skill for yourself; it appears only for you. Updating the skill to a new version keeps the connected account. After you delete the skill from your personal skills, Iskra deletes the saved Plaud access: if you install the skill again, you’ll need to connect your Plaud account again.

Connecting your Plaud account

1. Ask Iskra: “Connect Plaud”.

2. A form with a “Connect Plaud” link appears in the chat. Open it, sign in to your

Plaud account and allow access.

3. Return to the chat, check “I allowed access in Plaud” and click “Done”.

The skill itself checks that Plaud has granted access and names the connected account. If access wasn’t allowed in Plaud, Iskra asks you to open the link again. You can check the status at any time by asking “Is Plaud connected?”; Iskra shows whether the account is connected and whether Memory is ready for import.

To disconnect Plaud, ask “Disconnect Plaud”. Iskra deletes the saved access only in the current profile; files already saved to Memory stay where they are. Disconnecting in Iskra doesn’t revoke the permission inside your Plaud account; you revoke it in Plaud’s own settings.

Installing the skill for the organization makes it available to employees, but the Plaud account isn’t shared: each employee connects their own Plaud account, and the skill works only with that account’s recordings. If the organization deletes the skill, the previous access stops working, and after a new installation everyone needs to connect Plaud again. The saved access isn’t erased immediately in that case but is replaced on the new connection; to delete it in advance, ask “Disconnect Plaud” before the skill is deleted in the organization.

What the Plaud skill can do

Show the list of recordings and find a recording by title or dates. Iskra states which portion of the recordings the search covered: for older recordings, the list may be incomplete, and the skill says so directly rather than passing off an incomplete result as complete. In that case, narrow down the search period.

Read a Plaud summary or transcript. The transcript comes in pages with timestamps and speakers, so even a very long recording can be read in parts.

Break down a meeting: outcomes, decisions, tasks with owners and deadlines, open questions. The Plaud summary and Iskra’s conclusions are labeled separately.

Prepare a recording file in Markdown format, with the summary and transcript.

Send the meeting’s audio recording to the conversation files: ask “send the audio of this recording”. Iskra downloads the file itself and doesn’t send a link to the chat.

Import recordings to Memory, into a selected folder, resuming from the same place on the next run. How imported files behave is described in the section “Files imported by skills”.

Import new recordings regularly through “Plans”; see the example “Recurring Plaud import”.

Audio is not imported to Memory: Memory already holds the transcript made by Plaud, and Iskra would transcribe uploaded audio a second time, which takes longer and adds nothing. If you still need the audio in Memory, ask Iskra to save the file it sent to a folder.

Try, for example: “Show my Plaud recordings for this week”, “Break down the ‘Planning’ meeting from Plaud: decisions and tasks”, “Import Plaud recordings into the connected Memory folder”.

Plaud skill limitations

Each recording is saved as a separate file, up to ~3 MiB of text per recording. If the transcript is longer, the skill won’t save a truncated file as if it were complete; instead, it offers to read the transcript page by page.

If the transcript isn’t ready in Plaud yet, the skill says exactly that: the transcript isn’t ready yet. A breakdown can be based on the summary, and during import such a recording waits for its transcript and is processed next time.

The audio recording goes to the conversation files and lives with the conversation. A very large recording may not fit; then Iskra says so and offers to download the file in the Plaud app. During recurring import, audio is not downloaded: only the meeting text is saved.

Plaud stores and processes the recordings on its own hosting; Iskra receives them through the connection to your Plaud account.

Security and the import folder

The skill treats recording text as data, not commands: if the transcript contains a request such as “delete the files” or “forward this”, Iskra will relay it as meeting content but won’t carry it out.

To import to Memory, exactly one Memory folder with write access must be connected to the chat or the “Plans” entry, and all other folders must be read-only. If there are no folders with write access or there are several, the skill stops before reading recordings and asks you to change the set of connected folders. This keeps personal meeting recordings out of other folders, for example a shared department folder that happened to be connected with write access.

Overview

The skill connects Iskra to your amoCRM account: it helps you find and create companies, contacts and deals, fill in fields from documents, attach files, and work with tasks and notes. Before writing, it shows a draft for confirmation.

Overview

1. The amoCRM administrator opens amoMarket → the “⋯” menu in the top right →

“Create integration” → external integration. Fill in the name “Iskra”, the required description, for example “CRM assistant”, and the redirect link: https://cloud.iskrabot.ru for cloud Iskra or the address of your installation. Turn on access granting and save the integration.

2. In the created integration, open “Keys and scopes” → the “Long-lived

token” block → “Generate token”. Choose a validity period and confirm. The skill needs specifically the long-lived token: the integration’s secret key is a different value. The token is shown only once, so copy it right away to enter it in Iskra’s secure form.

3. In Iskra, open “Skills” → catalog → amoCRM and install the skill personally or for the

organization. On the card, click “Needs setup” (later, “Details” → “Settings”). If both options are available, choose the right “Connection” at the top of the window. In the “amoCRM connection” group, fill in:

Long-lived token: the value from the previous step.

If you are setting up the connection from the chat, click the “Configure skill secrets” button in the timeline. It opens the window; the button merely being prepared does not mean the connection has been saved. The subdomain, domain and connection level you have already specified are prefilled in the form.

4. Ask Iskra to show the pipelines to check the connection.

The shared organization connection is set up once by the owner or an administrator. It is available to employees who have been given access to this skill. A personal connection is available only to you: other organization members can’t see its settings or use it. It fully replaces the organization connection; the setup window warns about this choice and shows which connection is in effect. Learn more: Settings groups.

The saved token is hidden: you can replace it but not view it. You can revoke access in amoCRM. The API works with the permissions of the user who granted access; choosing an employee in Iskra sets the attribution of records, not additional permissions. Attachments also require the integration to have access to amoCRM files.

Overview

The first time you change data, the skill offers to pick yourself from the list of employees. This choice is remembered for the current connection and sets the author of records that support it and the default responsible user. You can assign a different responsible user separately. The “On behalf of” choice can be changed in the confirmation form. When the connection or token changes, the skill offers to choose the employee again.

Overview

Examples of requests:

“Find the company by the INN from this card and create it if it doesn’t exist yet”.

“Fill in the company details from the attached PDF”.

“Find information about this company online by its INN and show the sources”.

“Create a deal for the contract, link it to the company and attach the contract”.

“Link the existing deal to this company”.

“Show my overdue tasks” or “Summary of the main pipeline”.

Iskra reads the document text and uses available text recognition for scans. Illegible or contradictory details need clarification. Company search also checks Cyrillic and Latin spelling variants and abbreviations. If several cards match, choose the right one by name and details. The search can continue on the next pages.

If the first check by name found nothing and variants in the other alphabet haven’t been checked yet, the skill directs Iskra to repeat the search with a phonetic spelling. Changing the case or abbreviating in the same alphabet doesn’t count as such a check. It isn’t required for searches by INN, phone or email.

Re-creating a record with the same data is checked against the saved result of the operation and existing cards. Matches by name, phone, email and INN in structured details help avoid duplicates. If you need a separate object with matching data, explicitly ask for it to be created separately and check this action in the confirmation. If the amoCRM response is lost, Iskra first checks the result of the previous write.

A deal’s budget can be changed to zero. After creating or changing a card, Iskra re-reads it and tells you if it couldn’t confirm that the changes were saved.

Field lists and types are loaded from your amoCRM. Missing values can be collected in a single form and then checked in the draft. Employees are shown by name when the amoCRM user directory is available (it requires administrator rights). If the directory is unavailable, reading cards continues with a warning and IDs. A company card includes linked deals. When the company in a deal is replaced, the draft shows which link changes.

Overview

The owner or an administrator sets the rules once in the amoCRM settings:

1. In the organization profile, open “Skills” → amoCRM → settings.

2. Select “Organization”, then the “Team sales rules” block.

3. Enter the products separated by ;, a name template, for example {client} — {product}, and

the names of required fields separated by ;. Field names must match existing deal fields in amoCRM. If needed, specify the sales type and signing date fields.

4. Save the settings. The changes take effect the next time a deal is written.

You don’t need to re-enter the CRM address: new rules are automatically linked to the organization connection. If only a personal connection is set up, its address is used. The subdomain and domain can be filled in together with the rules in the same window. When you edit existing rules, their link is preserved.

Iskra determines the product from the deal’s content and builds the name from the linked company’s name and the product. Before a deal is created or changed, the company, product and required fields are checked. Values already filled in are taken into account on update; clearing a required field stops the write. The sales type is chosen from the CRM options based on context, and the signing date is taken from the document or the user’s confirmed answer.

The rules apply to the specified CRM account in this organization, including through a personal connection to it. They don’t give other employees access to the personal token. These rules don’t apply to a different CRM account. If no rules are configured, the deal uses the name and fields from the request. You don’t need a separate skill or a PDF with sales regulations for this.

Overview

The skill fills in existing amoCRM fields. If the company field list has no INN (taxpayer ID), KPP or OGRN, the amoCRM administrator needs to add them: open any company card → “Set up” → “Add field”. For separate identifiers, the “Text” type works, as it keeps leading zeros. Fields are created for all companies in the account, not for a single card. amoCRM’s guide to creating fields.

After saving, ask Iskra to re-read the company fields and fill in the details. You don’t need a toggle in the Iskra admin panel for this. Writing the details into a note is a separate option that Iskra will agree on with you.

Overview

Ask Iskra to attach a file uploaded to the chat to a specific deal. Various formats are supported, including PDF, images and office documents, up to 20 MB (20,000,000 bytes) per file. Before attaching, the skill shows the deal, file name and size. amoCRM’s own limits also apply.

The result of writing the deal and the result of the attachment are reported separately. If the file has uploaded but the link to the deal isn’t confirmed, Iskra reports this and can retry linking using the saved result. An uncertain server response means you need to check the card before creating the record again.

You can also ask Iskra to show a deal’s attachments, download a selected document to the current chat for reading or text recognition, remove an attachment from a deal, or upload a new version. Downloading and uploading support the same 20 MB limit; saving to the chat requires enough free space in the working folder.

Removing an attachment removes its link to the selected deal. The file itself in amoCRM and its links to other cards remain. A new version replaces the content of the shared file: before confirmation, Iskra shows the linked cards that will see the update. Re-attaching the same content to the same deal checks the version of the saved file. If the content has already been replaced, Iskra reports the mismatch; an unchanged file can be reused.

If the new version has uploaded but Iskra couldn’t save the operation confirmation, a repeated request checks the already uploaded version. You don’t need to send the content again for this.

Overview

For an existing task, you can change the text, the due date, and the responsible person, complete it with a result, or reopen it. In a regular note, you can correct the text: the old and new versions are shown before saving. System history entries cannot be edited this way.

You can attach a contact to an existing deal, remove its link to the deal, or make an already attached contact the primary one. Other links are kept; when a contact is unlinked, the contact card itself stays in the CRM.

Overview

The “Knowledge map” tab in the skill details accumulates knowledge about amoCRM pipelines and fields. You can build or update it with the “Explore” / “Rebuild map” button. For details, see “Connection knowledge map”.

Overview

Iskra connects to your 1C through the standard OData interface. It is available in any standard configuration built on managed forms (ERP, Trade Management, Integrated Automation, Enterprise Accounting, Small Business Management) and is enabled with built-in tools, without customization.

The skills showcase offers two options:

1C Connector (1c-connector): the database is published on the internet (an HTTPS address). Choose it if the 1C web client opens at a public address.

1C Connector via VPN (1c-connector-vpn): the database is accessible only inside the organization’s network. Iskra connects through your OpenVPN server. This option is for organizations: the VPN profile is set by the administrator.

Overview

1. Open “Skills” → the showcase catalog, find the “1C Connector” card (or “1C

Connector (VPN)”), and install it for yourself or for your organization.

2. On the skill card, click “Needs setup” (later, the same fields open from

“Details” → the “Settings” tab). The “Skill settings” window opens. If both options are available to you, your personal profile and the organization, choose the one you need in the “Connection” row at the top of the window. Below it is the “Connection to 1C” settings group:

Publication address: https://host/database, the address of the standard OData interface, including the database name.

Login: the integration account in 1C.

Password: write-only, like regular keys. You can only see whether a password is set, not view it; to replace it, enter a new one.

Allow writing: allows creating and changing records and posting documents.

Allow deletion: allows physical deletion. Both checkboxes are off by default, so the skill only reads.

3. You can set up the connection twice: personally for yourself and, if you are an owner or

administrator of the organization, at the organization level, where it applies to all employees. If both are filled in, the personal connection fully replaces the organization one. If you choose a personal connection over an already configured organization connection, Iskra warns you about this; next to the group you can see which connection is currently in effect. For more on choosing where to set it up, see “Skills” → “Settings groups (connections)”.

For the VPN option, the organization administrator also fills in the vpn_conn secret (in the same place, in the list of secrets under the settings group):

Overview

If something does not work, ask Iskra to “check the 1C connection”, and it will figure out what is wrong on its own: a wrong publication address (for example, an extra “tail” at the end of the address, in which case Iskra suggests a corrected version), a wrong login or password, insufficient permissions for the integration user, or the 1C service being unavailable at the moment. For any problem, and even if the connection is not set up at all, the settings form opens so you can fix the right field right away.

Overview

1. Publish the database on a web server (Designer → Administration →

Publish to web server) with “Publish standard OData interface” enabled.

2. Create a separate integration user (for example, “IskraBot”) and give it

the “Remote OData access” role plus read permissions (and write permissions, if writing is needed) for the objects you need.

3. Add the objects to the OData composition: Administration → Standard

OData interface settings → select the catalogs, documents, and registers you need. Without this step the skill will connect, but the object list will be empty.

4. After changing the composition, save and restart the publication.

Overview

Ask Iskra: “show warehouse stock from 1C”, “find items by name …”, “create a catalog item …” (with writing enabled). With its first call, the skill gets the list of available objects. If the list is empty, Iskra tells you that the OData composition has not been filled in (see step 3 above).

Both skills read lists page by page: by default, one page holds up to 100 records. If you need the whole list, explicitly ask to export it in full, and Iskra should keep reading the next pages. If reading stops because of an error or a limit, the reply should mark the export as partial and state the number of records received, the reason for stopping, and how to continue.

The skill card has a ready-made scenario, “Connect 1C and explore the data structure” (the “Scenarios” tab). It walks you through setting up the connection, checking it, and exploring the database objects, and saves the result to the knowledge map.

Overview

The skill details window has a “Knowledge map” tab with accumulated knowledge about the objects in your database: what each object and its fields mean, plus verified query examples. Iskra adds to the map as it works, including remembering a working approach it found after an error. You can rebuild or update the map with the “Explore” / “Rebuild map” button; for details, see “Skills” → “Connection knowledge map”.

Overview

Memory is a single place for personal files, projects, chats, and folders synced with your computer. There is no longer a separate “Knowledge base” section: after the update, its files appear in Memory automatically, and you do not need to move them manually. In the desktop app, a “Computer sync” action above the list opens the “Drive” page, where local folders are linked. A linked folder is already visible in Memory; no separate transfer is needed. Connecting such a folder to a chat or project and the limits of reverse sync are described in the “Folder sync” section.

Former “Knowledge base” folders are in the common list on the “Files” tab together with all the others; there is no separate tab or group for them. The filters above the list help you find them: the “My folders” chip shows your own folders, and the “Shared with me” chip shows folders shared with you. A specific folder or file is quicker to find by name with search. You do not need to upload these materials again; the migration requires no action from you. A hint at the top of the folder list says the same: “Learn more in Help” opens this help section, and “Got it” dismisses the hint for good.

Files and folders

In Memory, you can:

create regular and nested folders;

upload several files at once with the button or by dragging them into an open folder, and see the progress of each upload;

open, download, rename, and move files;

select several items and move them to the trash;

restore deleted items from the shared “Trash” tab within 30 days, if your permissions allow it.

A top-level folder you own has a “Rename” action in its “More” menu: enter a new name and click “Save”. Files, access, and connections to chats, projects, and collections are preserved. This also works for a pinned folder or a folder selected as a chat’s working folder.

Someone else’s top-level folder can be renamed and moved to the trash only by its owner; the “Manage access” permission alone does not allow this. These actions are not available for system-managed chat and project folders, synced folders, or read-only folders. An explanation is shown next to the action.

For nested folders and files, the “Rename or move” action separates the item name from the destination folder. Choose a folder in the list by navigating through the breadcrumbs, change the name if needed, and confirm the operation. Moving happens within the same Memory source. If you close a form you have already changed, Iskra first asks you to confirm that the entered data will be lost.

Clicking a folder name opens its contents. The other row actions are in the “More” menu: “Manage” opens agent access and the processing status, and the same menu has pinning, connecting to a project and a collection, moving, and deleting. Files also keep a quick preview next to the menu.

Each file and folder in the list shows the date and time it was created. When you open Memory for the first time without saved settings, “My folders” and the “Created” sort are selected, so new items are shown at the top. Your choice of filter and sorting is saved in this browser when you come back or reload, including the “All” filter. Previously saved settings remain in effect. Click “Name” to sort by name or “Created” to sort by date; clicking again reverses the order. If you have no personal folders yet, the “Show all” button opens the other folders available to you.

While a file is uploading, you can open other Memory folders, and the upload continues into the original folder. The full destination path is shown under the name of the file being uploaded; click it to open that folder without interrupting the upload.

Old and new office formats, including DOC, XLS, and DOCX, are uploaded as originals; text extraction runs separately after upload. If one file in a batch is not accepted, its row shows the specific reason, and the other files keep uploading. Once the cause is fixed, you can retry just that file.

When you upload a file and the folder already has a file with the same name, Iskra saves the new one next to it and adds a number to its name: for example, video (1).mp4, then video (2).mp4. The existing file is not replaced. Identical names in different folders do not conflict.

If you reload or close the page or leave the section, the browser warns you that the current upload will be interrupted. If you leave anyway, an “Upload interrupted” row appears when you come back: click “Choose file” and select the same local file. The upload starts over from the beginning and is saved to the folder chosen the first time.

Each file row shows its processing status. The original is available right after upload, even while Iskra is still extracting text, recognizing a scan, or transcribing a recording. If processing fails, the “Retry processing” action is available; you do not need to upload the original again.

Iskra automatically processes files in your Memory folders and project files. Files in a chat’s working folder are not processed automatically: they show the “Not processed” status. This applies both to the chat’s system-managed folder and to a Memory folder selected as the working folder: as soon as a folder becomes a working folder, new files in it, including those you upload through the “Memory” section, wait for the “Process” button. This is not an error: the original is available as usual, but the content of such a file is not included in content search and is not available to the agent as text. The “Process” button next to the status starts processing manually; this requires write access to the file. The same status may appear on files uploaded before one of the updates, and the button works for them the same way.

If a large or complex office document cannot be processed completely, Iskra automatically tries to extract its text in a simplified form. This text may lack formatting, images, charts, and embedded objects, but the original stays unchanged and remains available for viewing and download. If the text cannot be extracted, Memory keeps the file and its basic details, but content search and the agent will not be able to use the document text.

For a processed file, you can open “Extracted text”. This is a safe view of the current version: it is read-only, cannot be downloaded as a separate file, and does not show storage service data. Regular Markdown files created by you or the agent remain standalone editable files.

A folder as a knowledge base

You do not need to create or configure a knowledge base separately, let alone set up RAG or another search index for it. A regular Memory folder already works as a knowledge base: upload documents to it and connect the folder to a chat. From there, Iskra browses the folder structure, searches the files, and reads the relevant fragments on its own; with a large volume, it delegates the reading to subagents (see “Subagents within a single task” in the “Agent mode” section).

Files in Memory folders and project files are processed automatically, so their content becomes searchable right after processing. The exception is a chat’s working folder: its files wait for the “Process” button (see “Files and folders” above).

The agent sees a folder and refers to it by its regular name, for example “Contracts 2026”, just as you do in the interface.

Extracted text and the hidden mirror

For a processed non-text file, Iskra has access to a hidden view with the extracted text. It sits next to the original and is named <original>.text.md, for example contract.pdf.text.md. This is a read-only technical view: it is not shown in the “Memory” section, is not a separate file, and cannot be edited. You keep working with the original, and in its replies Iskra refers to the source document, not to the mirror.

The mirror contains the text but does not replace the original: the document’s layout, images, charts, and other visual relationships are not reproduced in it. When formatting or the content of an image matters, you need the original itself.

Lists: filters and pinned folders

The general Memory list (when no specific folder is open) can be narrowed with the filters above it: “All”, “My folders”, “Projects”, “Chats”, “Connected”, “Shared with me”. They correspond to folder categories. Filters are not shown inside an open folder.

The “My folders” chip shows only your own folders. Folders shared with you, whether directly by the owner, through your group, or through your organization, are collected in a separate “Shared with me” category and marked with a “Shared with you” badge in the list. This way, your own folder and someone else’s folder opened to you no longer look the same.

The selected sorting and filter are remembered: when you return to Memory in the same browser, you see the list the way you left it. If no folder matches the selected filter, instead of an empty screen you see “Nothing matches this filter” with a “Show all” button, which resets the filter to “All”.

Empty system-managed chat folders that do not have a single file yet are not shown in the list. As soon as the first file appears in such a folder, the folder shows up in the list automatically. Personal folders, project folders, and connected folders are always shown, even when empty, because you created or connected them explicitly.

Any top-level folder can be pinned: choose “Pin” in the “More” menu. Pinned folders are collected in a separate “Pinned” group above the rest of the list regardless of the origin filter applied, including an empty system-managed chat folder if it was pinned explicitly. To unpin a folder, use the same action. You can pin up to 50 folders per profile.

List rows are compact: the path is not shown under the name (it remains only in search results), the category is indicated by a small icon next to the name, and the creation date and file processing status are shown as before.

Board placement mark

A file placed on a board has a placement mark in Memory. In a list row the mark is short, with details in a tooltip; in the file card and the edit window it is detailed, with the tab status, the time of the version being shown, and a link to the board. In the file card, placements are collected in the “Board placement” block, where anyone who may change the source and to whom the server has disclosed the board itself is also offered “Refine in chat”.

The mark is also visible to someone who edits a shared source but does not own the board. The name of someone else’s board, its recipients, and the link secret are not included in the mark; only the fact of placement and the audience category remain. The mark itself does not grant access to the board: the board is still opened according to its audience.

Folder access and board access are different things. A folder member works with the folder’s files, while a board reader gets selected versions of selected files, not the folder and the rest of its contents. A file in the trash has no mark: readers cannot open the tab of a deleted source.

How to place a file, which audiences there are, and how “Follow the file” differs from a pinned version is explained in the “Boards” section.

Trash

On the “Memory” page, the “Trash” tab is shared across all sources available to you: your personal folder, projects, chats, and connected folders. For each deleted item, it shows the original path and source, the deletion date, and the time left until automatic permanent deletion. The list is reloaded each time you open the tab, so newly deleted items appear without refreshing the page. Files from “Drafts” sections are marked with their own source; if needed, you can narrow the list to just that source with the “Drafts” filter.

By default, the most recently deleted items are shown first. Click the “Name” or “Deleted” header to sort the whole trash by name or deletion date; clicking again reverses the order. On a narrow screen, the same options are above the list.

Once moved to the trash, a file or folder immediately disappears from the “Files” tab, from search results, and from the content available to the agent through connected folders. Right after deleting, you can click “Undo”; later, the item can be restored from the trash.

You can move a top-level folder you own to the trash as a whole, even an empty one. Before you confirm, Iskra warns that the folder and its contents are kept for 30 days and that its connections will be temporarily unavailable. Restoring brings back the same folder with its previous name, contents, access, and connections. Files and nested folders deleted earlier in a separate operation stay in the trash. Only the owner can see and restore a deleted top-level folder.

A regular file or folder can be restored within 30 days of each deletion. If you restore such an item and then move it to the trash again, the countdown starts over.

A file from “Drafts” follows a different rule: its total retention period is 30 days from the last change to its content. Viewing, downloading, renaming, moving, deleting, and restoring do not extend this period. For example, if a draft is deleted 10 days after its last change, it can be restored from the shared Trash only during the remaining 20 days. After restoring, the file returns to “Drafts” with its previous deadline; a new content version starts a new 30-day countdown.

When the applicable period expires, Iskra automatically starts permanent deletion. From that moment on, restoring is no longer possible. Until permanent cleanup is complete, the data keeps taking up storage space.

The trash can hold several separately deleted items with the same name and original path at the same time. They are shown as separate rows with their own deletion dates. Unlike a new upload, restoring does not replace an existing object or add a number to the name. If the original path is already taken, delete or rename the file or folder that is there and retry the restore. A conflict on a nested path blocks restoring the entire deleted folder.

A deleted folder is shown as a single row. Expand it to view its contents in read-only mode. Nested files can be opened and downloaded, but you cannot restore or permanently delete an individual item inside the folder: actions apply to the whole deleted folder.

The trash shows only items from sources you still have read access to. Restoring requires permission to change the source, and the “Delete permanently” and “Empty trash” actions require owner or administrator rights. If your access to a source is revoked, its items no longer appear in your trash.

To delete data before the period ends, select one or more items and click “Delete permanently”, or use “Empty trash”. Before an irreversible action, Iskra asks for confirmation. When you empty the trash, items you are not allowed to delete permanently remain in it. Permanent cleanup can take some time; while it runs, the “Deleting permanently” status is shown.

If permanent deletion fails, the row gets the “Deletion failed” status. If you have the rights, click “Retry deletion”. In this state, restoring is no longer available, because part of the data may already have been cleaned up. While an item is being restored or deleted, or its contents are loading, the button shows the action in progress and a progress indicator.

Search

Search on the “Files” tab starts with the “Search” button or the Enter key and covers all files and folders available to you outside the trash. It has three modes:

Name and path: searches the names of files and folders and parts of the path;

Content: looks through the extracted text of the current processed versions;

Everywhere: immediately shows matches by name and path, then adds what is found in the content.

While you edit the query, the current results stay on screen until the next run. Clearing the field immediately brings back the regular list. The applied query and search scope are saved in the page address, so you can copy a link to the results; when the link is opened, the search runs automatically. Changing the scope while a query is applied repeats the search immediately.

Results in the “Everywhere” and “Content” modes are sorted by relevance, so column sorting is not available for them. “Name and path” results can be sorted. The “Show in folder” action on a found item ends the search and opens the item’s location.

There is no persistent search index for content, so content search can take longer. You can stop it, and the results already found stay on screen. If some of the available folders could not be scanned for the moment, Iskra explicitly shows that the result is incomplete and offers to repeat the search.

In the folder pickers for a chat, project, or collection, search starts automatically after a short pause and tolerates one typo. Exact matches are shown first. Fuzzy search runs over a limited set of available folders, so if a folder was not found by an inexact query, refine or correct the name.

Chat working folder

By default, Iskra creates a system-managed folder for chat files. In a new personal chat, you can choose an existing top-level Memory folder as the working folder instead, as long as the chat has no messages and no files of its own. Message attachments and files created by the agent go to the working folder. In the request panel, its contents are shown in the “Chat files” area; when the folder has files, the Memory icon in the area header opens the same folder in the “Memory” section.

Until the first message is sent or the first file is uploaded or created in this chat, you can click the “Cancel workspace selection” cross next to the folder name to go back to choosing. The folder itself and all its files stay where they are. One folder can be the working folder for several chats: earlier files and files of other chats do not prevent you from cancelling the selection in an empty chat.

If the working folder itself has been moved to the trash, its link to the chat is kept. First restore the folder in Memory; after that, you can work with it again or cancel its selection in an empty chat.

After the first message or the chat’s own file, the working folder is locked in: to use a different folder, start a new chat. Deleting messages or moving files to the trash does not bring back the option to choose. If access to the selected folder has been revoked, you can cancel the selection in an empty chat; to continue a conversation already in progress, restore access or start a new chat with a different working folder.

Context connections are managed separately and are kept when you cancel the working folder selection. If you need context from somewhere else, connect a folder via “@ Folders” next to the input field. You can connect several folders; a single file

is added to a message as an attachment and does not give access to the other files in its source folder.

The folder pill in the input line shows the name of the connected folder or, if there are several, the name of the first one and a count of the rest, for example “Contracts 2026 +2”. The full list appears in a tooltip on hover. The “Connected folders” area of the request panel lists the conversation’s own connections, while folders inherited from a project are counted by the project card in the first row of the area; their list opens in the “Connected folders” window on the “Project” tab.

To move a file from a chat into your Memory folder, ask for it right in the conversation: “save this file to the AI-LABS folder”. The folder must be connected to the chat and open for writing; if it is not connected yet, Iskra will prompt you to connect it. The file itself is moved, not a summary of it, so documents, spreadsheets, and presentations are saved in their original format.

If your skill needs materials across different conversations, choose a separate Memory folder and link it in the skill card. The skills guide explains the difference between shared bundle resources, company materials, and files for a single task.

Files imported by skills

Some skills can save prepared files to Memory on their own; for example, the Plaud skill saves meeting recordings. The file is copied to the Memory folder in full, and its text does not pass through the model, so it is not distorted by paraphrasing. The same file size and storage space limits apply as for a regular upload. The destination folder must be connected to the chat or to a “Plans” entry with write permission.

A skill does not silently overwrite your files:

A repeated import updates the same file only if no one has changed it since the previous import. If the recording’s content has not changed, the file stays as it was.

A manually edited file. If you edited an imported file, the skill reports a conflict and keeps your version.

A moved or deleted file. Automatic import does not recreate it but notes this in its report.

The Plaud skill puts recordings into the Plaud/<account key> folder inside the selected Memory folder, with a separate subfolder for each Plaud account. The same place holds the import-state.json file, which stores the import progress so that the next run continues from where the previous one stopped. You do not need to edit this file. If a previous import was interrupted, then before updating a file in Memory the skill checks that the file is its own unmodified export; files you have edited stay as they are.

Collections

The second tab of the section is “Collections”, between “Files” and “Trash”. A collection is a personal set of several Memory folders that you connect to a chat or project in a single move, like a regular folder.

The tab lists your collections with the number of folders in each. The “Create collection” button opens a form with a name field. A collection can be renamed and deleted (deletion cannot be undone: there is no separate trash for collections yet). Deleting a collection disconnects it as context: chats it was connected to stop seeing it, and the folders inside it are not deleted. In a project’s “Connected” list, however, the row does not disappear by itself: it stays there named “Unavailable collection”, just like a deleted folder, and the project owner disconnects it manually.

Open a collection to see what it contains: a list of folders, any of which you can remove, and an “Add folder” button that opens the same folder picker as “@ Folders” in a chat. You can also add a folder to a collection right from the “Files” tab: the

folder menu has an “Add to collection →” action. If a folder in the collection is not yours but has been opened to you, for example within an organization, its row has a “Shared” label. This only indicates where the folder comes from; actual access is still checked separately (see below).

A collection does not expand permissions: access to each folder inside is still determined by that folder’s own access settings, and Iskra checks it again every time the collection is connected to a chat or project, not only when the folder was added to the collection. So a folder inside a collection may be inaccessible not only because access was revoked later (the folder was moved or deleted, or the permission was removed) but also because the viewer never had access in the first place, for example a project member to whom the collection owner did not open some of the folders. In both cases, Iskra simply does not connect that folder and does not show it to the agent; there is no separate label such as “Unavailable folder” for this. The other folders in the collection remain available as usual.

You can connect a collection to a chat via “@ Folders” and to a project via the “Memory” block on the project page: in both places, a separate “Collections” section appears next to the folder list. A connected collection behaves like a folder: you can narrow its mode to “Read only” or disconnect it entirely. In the list of connected items, a collection is always shown as a single row with its name, without expanding its contents. Only the collection’s owner can change its contents, and the changes are immediately visible to everyone whose chats or project it is connected to.

Automatic collections

Once an hour, if new folders or files have appeared in Memory, Iskra looks through their names on its own and may assemble a thematic collection from them. This complements manual collections rather than replacing them: both kinds live in the same list on the tab and are connected to a chat or project in the same way.

The analysis covers not only your own folders but also folders shared with you, including within an organization. Such a folder in a collection carries the same “Shared” label as when it is added manually.

A collection assembled by Iskra has an “AI” badge next to its name. Iskra can add new matching folders to such a collection, but it can remove only the folders it added itself; it will not touch a folder you added. Iskra cannot rename or delete a collection. Instead, if a collection has become empty or no longer makes sense, it appears in the “Recommendations” block above the list of collections as a card with the reason and two actions: “Accept” deletes the collection after the usual confirmation, and “Dismiss” leaves it unchanged. A notification tells you about new recommendations. Iskra creates recommendations only for collections it assembled itself: it will not suggest deleting a collection you assembled manually, even if it has become empty; only you can delete such a collection.

The card of any collection, including one you assembled manually, has a refill mode switch: “Automatic refill” or “Manual refill”. In manual mode, Iskra does not change what the collection contains, even if it created the collection itself. The same folder can belong to several collections at once.

When you undo an action yourself, by dismissing a recommendation, removing a folder added by Iskra, or deleting an automatic collection entirely, Iskra does not repeat it: for 30 days afterwards, it will not suggest the same folder, collection, or recommendation again.

You can turn off automatic collection building with the “Automatic collections” switch in the tab header: collections that have already been assembled stay in place but are no longer refilled. You can also exclude an individual top-level folder from the analysis: its “More” menu has a “Don’t analyze automatically” item.

Below, the collapsed “Journal” subsection lists the actions already taken: which collection was created, which folder was added or removed, which recommendation appeared, and how it was handled. The “Show more” button loads older entries.

Folder access

For a personal or connected (synced) top-level folder, the “Permissions and access” action is available in the “More” menu. Access can be managed by the folder owner or by an administrator.

The dialog shows who already has access and lets you add a new rule for the whole organization, a specific group, or an individual person (searched by name or email), with the “Read only”, “Write”, or “Manage access” level. You can delete a rule at any time; the folder itself stays where it is.

If the selected organization, group, or person already has access, the form shows it: an “Access is already granted” line with the current level appears under the selection, the current level is preselected in the list, and the button reads “Change level” and stays inactive until you change the level. This way, a previously granted “Write” level cannot be accidentally downgraded by clicking the button again.

The access level changes only from the state you see on the screen. If someone else has changed the rule in the meantime (for example, you opened the window in two tabs, or two people are setting up access at once), Iskra does not apply your change over theirs: it shows a message and refreshes the list so that you can decide based on the actual level.

For the recipient, things look different: a folder shared with you is still managed by its owner. If you do not have the “Manage access” permission, the settings window does not open; instead, the folder card shows the note “This folder is shared with you — its owner manages access”. Once you are granted this permission, the “Permissions and access” action works as usual.

System-managed chat folders and project folders do not have this window: their access is determined by access to the chat itself and by project membership, respectively, and an explanation is shown instead of the window. A Memory folder selected as a chat’s working folder remains your regular folder: it does have an access management window, and access to it is set by that folder’s rules, not by access to the chat.

Agent access

For a file or folder, the “Agent access” inspector offers two modes:

Allow changes removes the local restriction;

Read only prevents the agent from changing this resource and, for a folder, also applies to its contents.

Allowing changes does not by itself connect the resource to all chats or expand your permissions. The agent gets access only in a chat where the folder is connected, and can change it only when your permissions, the connection mode, and the resource setting all allow it at the same time. If you do not have permission to change the setting, the switch is unavailable, with an explanation of why.

If the permissions could not be loaded, you can retry the request in the inspector. A save error does not reset the selected mode: fix the cause and click “Retry saving”. After a successful operation, the inspector shows a confirmation.

If a folder is connected in “Read only” mode, Iskra will not promise to create, rename, or delete a file in it. To make changes, use the “Memory” section or switch the connection to “Allow changes”, if your

permissions and the resource setting allow it. A file created in the chat working folder does not count as saved to the connected Memory folder.

Deleting via chat

If you ask Iskra to delete files or a nested folder inside a connected Memory folder, it moves them to the trash. The connected folder itself stays. Name the folder and the files you need. Before deleting several files, Iskra states the folder path and what will be deleted. A follow-up such as “delete those too” refers to the previously selected folder and files; if the meaning is ambiguous, Iskra asks you to clarify the target. When only files are deleted, nested folders are kept.

Within 30 days, you can bring an item back from the “Trash” tab or ask Iskra to restore it. If the folder is connected read-only, Iskra explains the restriction and changes nothing. If the operation is unavailable or fails, Iskra keeps the selected target and tells you what was deleted and what remained. Items that have already been deleted can be restored.

Agent instructions

If a conversation about a folder produces a stable rule for working with its contents, for example “only signed documents go here” or “name files by contract number”, Iskra offers to record it in an iskra-agent.md file in the root of that folder. Iskra itself creates and edits the file based on what the user says; you do not need to write it by hand.

Such an instruction is shared by everyone who has access to the folder: whenever Iskra works with this folder, whether in a personal chat, a personal project chat, a scheduled task, or anywhere else the folder is connected, it reads the rule and follows it. Anyone with write permission to the folder can write or change iskra-agent.md, not only its owner. The instruction itself grants no additional permissions: it only describes how to work with the folder’s contents. If its text nevertheless demands sending data outside the folder, expanding anyone’s access, or bypassing the usual confirmations, Iskra is instructed to refuse such a demand and warn you about it in its reply. This is a rule the model follows, not a technical safeguard of the platform.

The full text of the instruction is sent to the model with every request in which this folder is connected. The private “Antonym” mode applies to it just as it does to the rest of the request: with Antonym enabled, the instruction is processed on Iskra’s servers. However, anyone the folder is open to can read the file, so iskra-agent.md must not contain passwords, access keys, or other secrets, only the rule for working with the folder.

You can also open and edit the file directly, without Iskra: in the “Memory” section, the iskra-agent.md row is marked with the “Agent instructions” icon, and the “More” menu has an “Edit text” action that opens an editor; changes are saved as a new version of the file. The same icon appears on the file in the chat file tree in the side panel. Next to it, you can see when the file was last changed.

In the request panel, the “Connected folders” area shows the connected folders and their modes: the “Read” / “Write” pill on a connection row switches the mode with one click. You can narrow a connection to “Read” or remove it with the cross; removing a connection does not delete the folder itself. A row’s mode is the intersection of your permissions on the folder and your choice for this conversation, so you can expand a connection to writing only where you have write permission on the folder itself; narrowing to “Read only” is always possible. If the owner expands your access later, the connected folder becomes writable on its own; you do not need to reconnect it, and any narrowing you made manually is kept. Folders connected before this update may stay in their previous mode: switch them once on the connection row. The set of conversation

skills is managed in the “Conversation skills” window, which opens with the “Skills” button in the bottom row of the panel; in a new chat, the “/ Skills” button is also available.

Overview

The skill connects Iskra to your Bitrix24 portal: searching leads, deals, contacts, and companies, a sales pipeline summary, the CRM timeline and activities, and Bitrix24 tasks. Records are always created and changed only after you confirm the write. The connection uses your portal’s inbound webhook, with no password and no OAuth app.

Overview

1. On the Bitrix24 portal (as an administrator): Developer resources → Other →

Inbound webhook → select the crm, user, and task permissions → create. Bitrix shows the webhook code only once, so copy it or the entire webhook URL right away.

2. In Iskra: Skills showcase → “Bitrix24” → add it to your organization (or personal

profile) → on the skill card, click “Needs setup” (later, the same fields open from “Details” → the “Settings” tab). The “Skill settings” window opens. If both options are available to you, your personal profile and the organization, choose the one you need in the “Connection” row at the top of the window. Below it is the “Connection to Bitrix24” settings group:

Portal address: for example, mycompany.bitrix24.ru (or the address of a self-hosted installation or the customer’s CNAME domain, if it is reachable over HTTPS from the internet);

Inbound webhook: you can paste either the short code like 1/abc123xyz (the part of the URL after /rest/) or the full webhook URL. The field is write-only, like regular keys: you cannot view the saved value, only replace it.

3. For an organization, the owner or an administrator fills in the connection once,

and all employees can work with it. A personal connection (if filled in) fully replaces the organization one. If you choose a personal connection over an already configured organization connection, Iskra warns you about this; next to the group you can see which one is currently in effect. For more on choosing where to set it up, see “Skills” → “Settings groups (connections)”.

The skill details window has a “Knowledge map” tab with accumulated knowledge about your portal’s custom fields and pipeline stages. Iskra adds to it as it works; you can build or update the map with the “Explore” / “Rebuild map” button. For details, see “Skills” → “Connection knowledge map”.

Overview

On the first action that requires an author (creating a lead, a task, and so on), the skill shows a list of the portal’s active employees and asks who you are. The choice is remembered: subsequent records are made on your behalf without asking again. For tasks, this choice sets the initial assignee and creator; before each task is created, you can check and change both roles. Dismissed employees are not included in the list.

Overview

Reading:

find a lead, deal, contact, or company by name, phone, or email;

show an entity card with all fields, including custom ones;

deal pipelines with their stages and lead statuses, the portal’s CRM mode (classic or simple);

a summary of open deals by pipeline: the number and total amount at each stage;

export of deals at a selected stage, including won and lost ones, with IDs, amounts, currencies, responsible persons, and dates;

CRM activities (calls, meetings, task activities): incomplete ones for an entity or an employee;

the latest timeline comments of an entity;

the list of the portal’s active employees;

descriptions of CRM entity fields, including custom ones;

Bitrix24 tasks: your own or those of a specific employee.

Writing (always with confirmation):

create a lead or deal, create a contact;

change CRM entity fields, including moving a deal to another stage and changing the responsible person;

add a comment to an entity’s timeline;

schedule a CRM activity with a deadline;

create a Bitrix24 task, choosing the assignee and the creator separately.

Overview

For example: “Export all deals at the ‘Deal won’ stage of the ‘Fit-out execution’ pipeline and prepare a dashboard”. Iskra determines the pipeline and stage, then reads the deals in pages of 50 records. You do not need to name specific deals for such an export. If several pipelines or stages have the same name, you will need to clarify which one you mean.

The report can include deal IDs and names, amounts and currencies, responsible persons, creation, modification, and closing dates, and the linked contact and company. Additional fields of an individual deal can be retrieved from its card. Amounts in different currencies are shown separately.

If the request was stopped or the portal returned an error, the part of the list received is not a complete export: the report must take into account the number of deals read and the reason for stopping. Reading continues from the ID of the last deal read, so deleting records that were already read does not shift the next page. Still, the export is not a snapshot: deals that move into the stage and have IDs that were already passed will not be included. For a final reconciliation, export the data at a time when the deals in the stage are not changing.

Overview

Any write to Bitrix24 first shows a draft (the title, fields, and responsible person) and is performed only after you confirm it. In the form, you can correct the values before sending, and the “On behalf of” field lets you switch the linked employee right there.

For a task, the form shows the assignee and the creator separately. You can change both before confirming. After the task is created, the skill reads it back from Bitrix24 and tells you if the portal saved different roles or did not allow them to be checked. The task itself has already been created in that case, so you do not need to repeat the command.

Overview

The skill’s ready-made scenarios appear on the “Scenarios” tab of the skill card, marked “from author”, and start with the “Open chat with scenario” button without any additional setup:

Connect Bitrix24: setting up the inbound webhook and checking the connection;

Morning sales summary: pipeline status, amounts by stage, upcoming activities and tasks;

Qualifying a new inquiry: checking for duplicates by phone/email and creating a lead or a deal with a contact;

Preparing for a deal meeting: the deal card, communication history, open activities, and contacts in one brief;

Reviewing stalled deals: a pipeline summary, plus a comment and a next step for deals with no movement;

Exporting deals at a stage: a list of deals at the selected stage for a report or dashboard, including won ones;

Call results to CRM: a timeline comment, the next activity, and, if the deal moved forward, a stage change.

Ready-made scenarios cannot be renamed, edited, or deleted: they are the author’s bundle, shared by everyone who has installed the skill. You can save your own scenario based on a similar task separately, right from the chat.

Overview

The Bitrix24 REST API is available only on the portal’s commercial plans; on the free plan, the skill cannot connect.

Bitrix24 limits the rate of REST API requests. After a short burst, the skill retries once on its own; under a stricter limit, it replies “try again later”.

The pipeline summary covers up to 1,000 open deals; if there are more deals, the reply explicitly marks the coverage as incomplete.

The list of active employees is limited to 250. If the portal has more employees and you are not in the list when choosing “who you are”, contact the portal administrator.

Export of a selected stage runs in pages of 50 deals; the summary’s limit of 1,000 open deals does not apply to it. If the export stops or hits an error, the partially read list remains incomplete.

The skill cannot connect to self-hosted Bitrix24 installations that have no public DNS name or whose address is only on an internal network (a private IP): it reaches the portal over HTTPS through the platform’s network policy, which does not let internal addresses through under any settings.

Overview

Error — What it means — What to do

webhook_invalid

What it means: the webhook was deleted on the portal

or the code was entered incorrectly.

What to do: recreate the inbound webhook on the portal

and update the skill settings.

What it means: the webhook lacks permissions for the called method.

What to do: in the webhook settings on the portal, add the missing access permission

(crm, user, task).

Error — What it means — What to do

What it means: the REST API is not available on the portal’s current plan. What to do: switch to a commercial Bitrix24 plan.

not_configured

What it means: the skill is not connected.

What to do: fill in the “Connection to Bitrix24” group in the skill settings.

What it means: Bitrix24 has limited the request rate. What to do: retry the action after a while.

Overview

The webhook is kept in the platform’s protected storage and is not shown to anyone, including the agent. The webhook has the same permissions as the employee who created it, so create the webhook under an account with the minimum necessary permissions and revoke it on the portal when that employee leaves. The skill accesses only the portal address specified in the settings.

Overview

“Plans” is the section with your personal calendar and task list in Iskra: everyday to-dos and meetings sit alongside tasks that Iskra itself performs on your command. The section replaced the former “Schedule”. Recurring agent runs have not gone anywhere, but they now live next to your regular planner instead of separately from it.

The section has three tabs: Pulse, Tasks, and Calendar.

Overview

They are easy to confuse, but they are different things:

“Plans” (this section) is your own calendar and task list inside Iskra, including recurring and one-off agent runs. It lives in the platform itself; you do not need to connect anything.

The “Calendar” skill handles one-off work with your external calendar (Yandex, Mail.ru, iCloud, Nextcloud, a corporate CalDAV server) or quickly creates an event in Google Calendar: the agent goes there, checks, and writes. For details, see the “Calendar” skill section.

Calendar and task sync is a continuous exchange: events and tasks travel between “Plans” and an external calendar in both directions on their own, without a request in chat. You set it up once in Settings → “Calendars and tasks”; for details, see the “Calendar and task sync” section.

If you have an external calendar set up, “schedule a meeting” may go to it, while “create a task” goes to “Plans”; if the wording is ambiguous, Iskra asks where exactly to record it.

Overview

Pulse opens first and shows what has happened and what is coming up:

Happened: cards for events that have already occurred, such as a reminder (with “Done” / “Snooze” / “Open chat” buttons), an Iskra wake-up (a link to the chat with the result), or a skip with an honest reason (for example, “Skipped after a long silence” if the platform has not seen any activity from you for a long time, “Skipped: the reminder went stale” if the reminder did not arrive in time, or “Iskra meant to come back to this entry, but the chat was busy”; see “What happens when an entry fires”). Upcoming: the nearest tasks and events for the next few days, such as “📅 Tomorrow: meeting at 15:00”.

The “Snooze” button on a reminder offers preset options: +15 minutes, +1 hour, and tomorrow at the same time. Only that particular firing is snoozed, not the whole entry: for a recurring task, later occurrences stay where they are. The card is then marked “Snoozed”, and the entry itself on the “Tasks” tab shows a “Snoozed until …” line and a “Don’t snooze” button, which cancels the snooze and restores the original time.

If the selected period has too many recurrences, the feed does not show all of them and says so openly with the line “Not every occurrence of the period is shown”; narrow the period to see the rest.

Pulse is the shared history of firings: an individual entry has no run list of its own, and all of its runs and reminders are shown right here.

Self-tasks that Iskra sets for itself (see below) always appear in Pulse when they fire, regardless of the “Iskra’s internal plan” toggle on the “Tasks” tab.

Overview

A list of your tasks and events, with or without a due date. The “Create item” button at the top of the section opens the form (see “Creating an entry” below).

The period filters follow the order in which people usually ask about a list: Upcoming, Today, Overdue, No due date, Completed, and the general All view. The tab opens on Upcoming, that is, what still needs to be done. A task without a due date lives in the No due date group, not among the upcoming ones.

Separately, there are role filters: Assigned to me and Assigned by me (see “Assignments to colleagues”).

The “Iskra’s internal plan” toggle shows entries that Iskra sets for itself, for example the promise “I’ll get back to this in 5 minutes” made in a chat. They are hidden by default so that the list stays your own plan instead of being mixed with the agent’s internal bookkeeping. Hiding applies only to upcoming entries: every agent wake-up that has fired is always visible in Pulse, and its usage counts toward the profile’s overall statistics. Each run works in its own conversation and appears in “Top chats” on the “Statistics” tab alongside regular chats. There is no separate “self-task usage” line in the statistics.

Each entry in the list has a “Pause” / “Resume” toggle (it suspends firings without deleting the entry itself), as well as “Edit” and “Delete”. For a task that has already fired, “Done” and “Cancel” are available; for a recurring entry, both actions first ask about the scope: only this occurrence or the whole series.

The “Delete” button is also in the open entry itself, at the bottom left of the form, including when you open the entry from the calendar. It asks for confirmation: the entry’s reminders and runs are removed together with it.

Overview

Tasks and events share one entry form:

Entry type: Task (a single “Due” field) or Event (“Starts” and “Ends”). For a new event, the end is set one hour after the start and moves along with it; a later end you set manually stays as you set it. An event without an end is allowed: just clear the field. Besides the due date, a task has a separate optional “Work starts” field: when you plan to start on it, not when it must be done. The start cannot be later than the due date (equal is fine: a task started exactly at its due time is valid); reminders and firing are still tied to the due date, and a task without a due date will not fire even if a start is set.

Title and Note: a short text and, if needed, a longer description.

Project: you can link the entry to one of your projects; each agent run then uses that project’s instructions and files as they are at the moment the entry fires.

All day: marks the entry as not tied to a specific time of day.

Time zone: taken from your profile by default or, if it is not set there, from your device settings; you can pick a different one right in the form. This time zone determines when the entry fires. Times and dates, however, are shown everywhere (in the list, Pulse and the calendar) in your profile’s time zone, the same one the form fills in for a new entry: if you open Plans from another device or while traveling, you see your usual hours, not local time.

You set your own time zone once in Settings → Profile → Time zone (how to do it); it is usually already filled in with the time zone of the device you signed up or signed in from. Iskra also uses it to interpret “tomorrow at 9” and “in an hour” when you describe a task in chat. Until the time zone is set, Iskra asks for it before setting a due date in local time, and until then entry times are shown in UTC. Depending on your region, that can be a difference of several hours, so set your time zone before your first task with a time.

Entries a colleague assigned to you and meetings you were invited to are not yours to edit: for them you get assignee and participant actions, not editing of someone else’s plan.

Overview

The remaining fields of the same form:

Repeat: “Manual only” (no repeat) or one of the presets: Daily, Weekdays, Weekly (with a choice of weekday) or Monthly (with a choice of day of the month). Entries migrated from the old “Schedule” or created by the agent with a more complex rule may carry a rule that does not match any preset; the form shows it as a plain-language description and does not force you to convert it to the nearest preset.

Reminders: you can turn on one or more: 15 minutes, 1 hour, 1 day or 1 week before the due time.

Action: what happens when the entry fires (see the next section).

Push notification and Email notification: separate switches that choose which channels deliver a reminder or a run report.

Assignee (for a task) and Participants (for an event): you can assign work to a colleague or invite people to a meeting right here, not only by asking Iskra in a conversation. Both fields are available to the organization’s owner and administrators: in a personal profile and for a regular organization member, the form does not have these fields (see “Assignments to colleagues”).

Overview

More fields of the same form. Place and call link are shared by tasks and events; external attendees exist only for events.

Place: where it happens, such as an address, a meeting room or a venue name. It is shown on the entry card itself.

Call link: the video meeting address; on the card it opens with a single click. Only a full address of the form https://… is accepted.

External attendees (events only): guests without an Iskra profile, invited by email; enter the address and, if you know it, the name. Do not invite a colleague from your own organization here; use the “Participants” field above instead, so they see the meeting in their “Plans” and get its reminders. An external guest can be marked as optional: their attendance is welcome but not required. The response to the invitation (“Awaiting response”, “Accepted”, “Tentative” or “Declined”) is shown on the guest’s card, but you cannot set it yourself: it comes from the attendee’s calendar and is not set in Iskra itself. The platform administrator sets how many external attendees fit into one meeting.

Overview

Four more fields sit in the form’s “More” section, which is collapsed by default, together with the familiar “Project” (see “Creating an entry” above) and the notification method. The section expands with the “Show more” button; as soon as at least one field in it is filled in, the button also shows a count: “Show more · N”.

Importance: three levels, Low, Normal and High. Normal is set on almost every entry and is not highlighted in the list; High is marked with “!” next to the title in the “Tasks” list and in all calendar views (month, week, day). Low importance has no icon of its own.

Tags: your own labels for the entry, up to 10 per entry and up to 32 characters each. Extra spaces are trimmed, and two spellings of the same tag that differ only in case (“Work” and “work”) merge into one; the one you typed first is kept. A tag ends with a comma or the Enter key, so you can paste a whole list like “work, reports, home”. As you type, the field suggests tags from your profile’s shared vocabulary, that is, the ones already used on your other entries.

Busy status: the “Does not block time” switch; turn it on if the entry should not count as busy time.

Privacy: the “Hide details in the external calendar” switch marks the entry as private.

The filter that uses the same tags to narrow down entries across the whole section is described in “Filter by tags” below.

Overview

The “Action” field in the form offers one of three options:

No action: a regular to-do or calendar item; nothing is sent;

Remind: a push notification and/or an email at the due time, with no LLM involved;

Run Iskra: Iskra does the work itself at that moment (see “What happens when an entry fires”). This option adds the “What Iskra should do” field, where you describe the task the way you would explain it to a colleague, plus the “Memory folders” and “Skills” blocks: the connections Iskra will see in this particular run, regardless of what is connected in the chat when the entry fires.

The action type also determines what costs money: runs consume tokens and depend on your plan, while reminders and items with no action do not.

Overview

Scheduled Iskra runs are a paid feature: whether recurring and deferred agent runs are available depends on your plan. Regular reminders and tasks with no action work on any plan, including the free one.

The form will not accept a repeat that is too frequent: the minimum interval differs for an Iskra run and for a simple reminder, with runs allowed less often and reminders more often. This protects against an accidental storm of notifications and requests; it is not a plan restriction.

Besides the interval, there are count caps: how many entries one profile can have, how many of them can have an Iskra run, how many self-tasks Iskra creates for itself, how many reminders are delivered per day, and how many assignments you give out per day. The exact values are set by your platform administrator and differ between installations.

The number of internal colleagues in one meeting has no fixed cap. As before, you can invite only available members of your own organization, and only if you have organizer rights. The limit on external guests invited by email is set separately and still applies.

When you hit a cap, the form does not refuse silently: a line above the fields explains which limit has been reached and what to do about it. For example, “Limit of entries with an Iskra run reached: pause the ones you no longer need or move to a higher plan”, “The daily assignment limit is used up: continue tomorrow” or “The assignee has too many assignments: wait until they close some”. Your entry is not lost: the form stays open with your data, so you can adjust it and save again.

Overview

The form is not the only way. You can simply ask Iskra in a conversation: “remind me tomorrow at 10 to call the bank”, “come back to this in 5 minutes”, “every Monday, put together a summary of my email”. Iskra creates the entry itself and shows it in the chat as a card: what exactly is scheduled and for when, with a “Cancel” button, so you can change your mind right away without opening the section. A cancelled card stays in the conversation, marked “Cancelled”.

The same fields as in the section’s form (place, call link, external attendees by email, tags, importance, busy status, privacy and, for a task, a separate work start) can be given right in the conversation: Iskra passes them through the same action it uses to create an entry from the form. The card Iskra posts in the chat shows what sets the meeting apart from an ordinary planner line: the place, the number of invitees (the addresses themselves stay in the entry, not in the conversation), tags, and importance if it is higher than normal.

An assignment to a colleague works the same way: name the person or give their email address (“assign Ivan Petrov to prepare the report by Friday”), and Iskra finds them among your organization colleagues. If there are several matches or the person is not found, Iskra asks you instead of picking at random.

Overview

The “Run Iskra” action follows one of two paths, depending on where the task originated:

Wake-up in the original chat: if the task came from a conversation (for example, you asked “come back to this in 5 minutes” or “send this email tomorrow morning”), Iskra wakes up in the same chat and continues working with the context that was already there. If the chat is busy at that moment, for example with another request of yours, Iskra

waits and tries again later. If the chat stays busy for too long, Iskra will not move the work to a new chat without your knowledge: instead you get a reminder like “I was going to come back to ‘…’, but the chat was busy” with a link to that chat, so you can continue yourself. Iskra’s reply after waking up is set off in the conversation by a divider labeled “Iskra came back as planned at 22:47” with a link to the entry itself: from that point on it is planned work rather than a continuation of your conversation, and you can see what time Iskra started on it. Runs that took place before this label was introduced show no time in it.

New session: if the task was created in the “Plans” section or migrated from the old “Schedule”, firing opens a new chat and does the work there, with the Memory folders and skills selected in the entry form. A link to the result arrives as a notification and, if enabled, by email.

Wherever the “Run Iskra” action is available in “Plans”, in both cases the run happens without your involvement: if the task wording allows different interpretations (“clean out unimportant emails”), Iskra picks a reasonable option and names it in the report, and it does not silently perform irreversible or questionable actions: it proposes them as a separate item that you only need to confirm with a reply in the same chat.

Separately, an entry with a run has a “Run now” button: it starts the run immediately, and you see the response in real time, as in a regular chat.

Overview

If you have the Plaud skill installed, you can save new meeting recordings to Memory on a schedule.

1. Connect Plaud in a chat beforehand: ask “Connect Plaud”. An automatic

run does not show forms, so you cannot sign in to Plaud in the middle of a run. It is convenient to do the first import right there in the chat: that way you immediately see which folder is selected and what the report looks like.

2. Create an entry in “Plans” and choose “Run Iskra” in the “Action” field.

3. In the “What Iskra should do” field, write, for example: “Import new

Plaud recordings to Memory and send me a report”.

4. In the “Skills” block, select Plaud.

5. In the “Memory folders” block, connect exactly one folder for the recordings and leave it in

“Read and write” mode. Switch all other connected folders to “Read only”; otherwise the skill will stop and ask you to change the set of folders.

6. Set “Repeat”, for example to daily, and save the entry.

The run is performed on behalf of the entry owner and uses the owner’s Plaud connection. The run report lists new recordings, updated ones and those awaiting transcription. If there is no access to Plaud (for example, you disconnected Plaud), the report ends with the note “Connection required”. Reconnect Plaud in a chat: the import progress is saved, and the next run picks up where it left off. A missed run is caught up by the next one, so recordings from the missed period are not lost.

The recurring run checks for new recordings. If an older recording was corrected in Plaud, ask for a full re-import in a chat: “Re-read all Plaud recordings and update them in Memory”.

Overview

Your events and tasks with a due date in three views: Month, Week and Day. The view switcher and the arrows for paging through periods are at the top; the “Today” button takes you back to the current date.

Month: a grid of cells; a day’s cell shows the first entries, and the rest collapse into “N more”. Clicking a cell opens the entry creation form for that day.

Week and Day: an hourly timeline; an entry sits at its hour and its height matches its duration, while “all day” entries are gathered in a row above the timeline. Entries that overlap in time share the width of the day, so none hides behind another. A red line shows the current time. Clicking an empty spot creates an entry at the hour you pointed to (to the nearest half hour), so you don’t have to type the time.

Overview

If you connected a calendar on the Settings → “Calendars and tasks” page, entries from that calendar appear in “Plans”, and yours are sent there. The connection and everything related to it are described in “Calendar and task sync”; in “Plans” itself, it looks like this:

on the “Calendar” tab, a “Connect a calendar” button with a calendar icon sits on its own row above the view switcher; it leads to the connections page. While the list of connections is still loading, and when connections exist, the link is called “Calendars and tasks”. When connections exist, a pill next to it shows the overall sync status: “Syncing”, “Paused”, “Error” or “Sign in again”;

the card of an entry that came from a calendar says “from the calendar ‘…’”, and the card of an invitation shows who the organizer is and your response to the invitation;

an entry that came from a read-only calendar (such a calendar does not let Iskra write to it; in the connection’s calendar list it is marked “read-only”) is edited in that calendar itself: the time, title, place, reminders and external attendees are changed there, while in Iskra the entry keeps its project, tags, internal participants, notifications and Iskra’s work. Entries from a regular calendar are edited in Iskra like your own, and the changes are sent out to the calendar;

the entry form shows where the entry will go (“Will appear in: …”) or what it already syncs with; the “stop syncing” link unlinks a single entry: it stays in Iskra, and the item in the external calendar does not disappear;

if syncing breaks, a card in “Pulse” tells you so, with a link to the settings, for example “you need to sign in again”.

Entries from a calendar do not count toward your profile’s entry quota.

Overview

Each tab (Pulse, Tasks and Calendar) shows tag chips above its list, feed or calendar: click a tag to keep only the entries that have it, and click it again to clear the filter. You can select several at once; then entries with at least one of the selected tags remain. By default, up to eight of the most frequent tags in your vocabulary are shown; the “Show more · N” button reveals the rest, and “Clear tags” (visible while at least one tag is selected) clears the filter entirely.

Each tab has its own selection: if you filter “Tasks” by the “Work” tag, you won’t see the same filter on “Calendar”, which has its own. The selection survives a page reload and is kept in the page URL, so you can save or share a link to the filtered view.

A tag in the filter is spelled exactly as on the entry itself: case matters, and “Work” and “work” are two different tags in the vocabulary and in the chips (unlike within a single entry, where such duplicates merge as you type; see “Importance, tags, busy status and privacy” above).

Overview

The owner or an administrator of an organization can assign a task to a colleague in the same organization in two ways. First, with the “Assignee” field in the entry form (“Participants” for an event). Second, by asking Iskra in chat to “distribute” the action items of meeting minutes among assignees: Iskra prepares a draft of who gets which task and creates all the assignments with a single confirmation. Other organization members and personal profiles do not have these fields in the form.

A meeting with several colleagues is set up in a similar way: they become participants of the event (this is not an assignment: an event has no assignee) and are notified about the meeting itself and its reminders. The platform administrator sets how many people fit into one meeting.

In the first version, assignments are one-off only, with no repeat: you cannot create a recurring assignment for a colleague yet.

Overview

The assignee sees the assignment on the “Tasks” tab under the “Assigned to me” label: the title, note, due date and the name of the person who assigned it, plus, if set, the place, tags and importance above normal. The card itself also shows the reminders for this task and the snooze, if the next reminder has been snoozed. A meeting participant sees the same fields on their card. The chat where the assignment originated, its settings and the run setup are not visible to the assignee: they get only what relates to the task itself. Likewise, a meeting participant cannot see the organizer’s chat.

The assignee can decline the assignment and, if they wish, briefly explain why. The reason is optional (the right to decline does not depend on being willing to explain it), but if given, it is sent to the assigner along with the decline notification and stays in the entry itself: both sides see it in the row of the declined assignment.

Overview

Reminders and run reports arrive through several channels at once: in Iskra itself, as a push notification on your phone and, if enabled in the form, by email.

A notification takes you straight to the point: it opens “Plans” at the relevant Pulse card or at the entry itself, and if the entry belongs to another of your profiles, Iskra first switches the working context so the entry is visible.

Mobile push notifications are currently delivered via Google’s notification service (FCM). If your device has no Google services (for example, a phone where the app was installed from RuStore without Google Play Services), push notifications may not arrive; the entry itself, the “Pulse” feed and, with email notification on, the email still work as usual. We state this limitation openly rather than hide it.

Overview

The “Schedule” section no longer exists: all previously created recurring tasks were automatically moved to “Plans” and keep running on the same schedule; they appear on the “Tasks” tab just like newly created ones. Old links to the “Schedule” section open “Plans”.

The history of past runs was moved too, but it now lives in the shared feed: an entry has no separate run list like the one it had in “Schedule”; runs are shown in Pulse, and each result opens via a link to its conversation.

A few tasks with an unusual schedule may have been moved over in a paused state: their repeat rule does not match the form’s presets, so the decision was left to you. Such a task appears in the list with its time; choose a suitable repeat and resume it, and it will run again.

Overview

The skill helps you add a ready event to Google Calendar in one quick step, and it also connects Iskra to your personal calendar over CalDAV: it shows your plan for the day or week, finds meetings and free slots, and creates, reschedules and cancels meetings. The full connection works with Yandex Calendar, Mail.ru, iCloud, Nextcloud and corporate CalDAV servers. Credentials never leave the platform: Iskra handles all communication with the connected calendar server, and the agent never sees the password.

The skill is for one-off actions on request: “show my plan for tomorrow”, “reschedule the meeting”. If you want events and tasks to flow between Iskra and your calendar continuously and automatically, that is a separate setting, “Calendars and tasks” in Settings; see “Calendar and task sync”. The two do not interfere with each other: you can keep the skill or not install it at all.

Their connections are also different, which matters at first setup. In the “Calendars and tasks” wizard, you pick the card for your service, and it already knows the server address, so there is no address field. The skill has no cards: here the server address is a regular form field, which you fill in manually for email on an organization domain. Details below.

Overview

Name the event and give the date, time and duration, for example: “Add a strength workout to Google Calendar on August 18 at 15:00 for an hour”. Iskra immediately prepares a link to a pre-filled event form. Open it, check the details and click “Save” in Google Calendar.

You can give a duration (“for an hour”) or an exact end time: Iskra calculates the end of the event itself and correctly handles events that run past midnight into the next day.

This action needs no app password and no calendar connection. The event is not created until you save it in the Google form that opens. If you explicitly named a time zone, Iskra passes it to the form; without one, the time is interpreted in your Google account’s time zone. If you did not name a target calendar, Iskra asks one short question so you can choose between Google Calendar and a calendar you have already connected.

Installing the skill for quick Google events does not open the connection form. CalDAV settings are needed later only for reading your calendar, finding free slots or changing meetings through Iskra.

The quick Google route prepares one-off events, either timed or all-day. For a request with a repeat, Iskra explains this limitation and does not turn the series into a single occurrence. A link is prepared only after you explicitly decide to switch to a fully specified one-off event; recurrence/RRULE is not added to it.

Overview

You need the full connection when you want to read your calendar, find free slots or change meetings directly through Iskra. It is set up with an app password:

a separate password that you create with your provider specifically for third-party apps (not your main email password).

App password: Yandex, Mail.ru, iCloud, Nextcloud

Step 1. Create an app password with your provider.

Yandex: Yandex ID → “Security” → “App passwords”. Yandex keeps separate passwords for each access type, and the calendar needs the “CalDAV for Calendar” category specifically. A mail password will not open the calendar: this is a separate password, so create a new one in the right category. For a Yandex 360 mailbox on an organization domain, everything is the same: the same Yandex ID and the same password category. Mail.ru: Mail settings → “Security” → “Passwords for external apps”. iCloud: appleid.apple.com → “Security” → app-specific password (two-factor authentication must be turned on for your Apple ID).

Nextcloud / corporate server: your regular password or an app password (if your administrator has enabled them); get the CalDAV server address from your administrator.

Server address, time zone and working hours

Step 2. Fill in the skill settings in Iskra. Skill catalog → “Calendar” → add to your personal profile → on the skill card, click “Needs setup” (later, “Details” → the “Settings” tab opens the same fields). In the “Skill settings” window, the “Calendar connection” settings group opens:

Login (email address): the full address, for example ivan@yandex.ru . For Yandex 360 email on an organization domain, use the address with that domain, for example ivan@corp.ru ; for iCloud, the login remains your Apple ID even if your mail arrives at a custom-domain address;

Server address: if your email is on the service’s public domain ( @yandex.ru , @ya.ru ,

leave it empty: Iskra determines the server from your email address. If your email is on an organization domain (Yandex 360, Mail.ru for Business, iCloud+ with a custom domain),

or https://caldav.icloud.com : the calendar server usually cannot be found from the company domain. For Nextcloud or a corporate server, enter it explicitly, for example cloud.company.ru ; you do not need to specify the exact CalDAV path, because Iskra tries the standard options ( /.well-known/caldav , /caldav/ , /dav/ ) at the address you entered. If you enter http://… , Iskra replaces it with https:// itself: the connection will not run over an unsecured protocol;

Time zone: can be left empty, since it is usually already known from your profile (the browser reports it when you sign in). Fill it in manually only if Iskra asks;

Working hours: used when searching for free slots; the default is 09:00–18:00 on weekdays;

App password: paste the password you created in step 1. Like regular keys, the field is write-only: you cannot view the saved value, only replace it.

Step 3. The connection is personal: the calendar is set up for your profile, not for the whole organization, and the agent works with your personal calendar.

Enter connection details only through the skill settings form. Do not paste the app password into your chat with the agent.

Overview

Quick event in Google Calendar: a pre-filled form with the title, time, place and description, with no connection setup; you complete the save in Google Calendar yourself.

Plan for the day or week in a connected CalDAV calendar: events for the period across all your calendars at once, with conflicts (meetings that overlap in time) and “all day” events marked. Iskra names the calendars the answer was compiled from.

Finding a meeting in a connected CalDAV calendar: search for a meeting by title within a period; letter case does not matter, including non-Latin scripts (“Standup” and “standup” match).

Free slots in a connected CalDAV calendar: free time of the required length within your working hours, taking into account busy time across all calendars.

Creating a meeting in a connected CalDAV calendar: with a place, attendees, a description, a reminder (“remind before”, from 10 minutes to 1 day) and a repeat (daily, weekdays, weekly, monthly, yearly) with an end date.

Rescheduling and editing a CalDAV meeting: time, title, place, reminder, description.

Cancelling a CalDAV meeting.

For a recurring meeting, when rescheduling, editing or cancelling, Iskra asks about the scope: only this occurrence or the whole series.

Overview

In a connected CalDAV calendar, creating, rescheduling and editing a meeting go through a draft form: Iskra shows all the fields of the upcoming meeting, and you can adjust them and confirm with the form button (“Create a meeting” / “Save”). No separate confirmation in the conversation is needed: clicking the button in the form is the confirmation. Iskra cancels a meeting only after you explicitly agree in a chat message.

For Google Calendar, Iskra opens Google’s own form. The event appears in the calendar only after you save it in that form.

Overview

You can add meeting attendees, and they will be included in the event, but no invitations are sent to them: nobody receives an invitation email. Iskra warns you about this when creating the meeting. You can send invitations separately, for example by email through the “Mail” skill.

You also cannot respond to someone else’s invitation (accept or decline a meeting you were invited to) through the skill; that is done in the provider’s calendar itself.

Overview

These are two different mechanisms, and their names make them easy to confuse:

“Plans” (an Iskra section) is your own calendar and task list inside the platform, including recurring agent runs: tasks that Iskra performs on its own at set times. This section is not linked to your external personal meeting calendar. For details, see the “Plans” section.

“Calendar” (this skill) is about events in your external personal calendar: a quick pre-filled action for Google Calendar, or reading and editing meetings on a connected server (Yandex, Mail.ru, iCloud, Nextcloud). It does not run the agent on a schedule.

Overview

Invitations are not sent, and you cannot respond to someone else’s invitation (see above).

The “this and following” scope for recurring meetings is not supported: only a single occurrence or the entire series.

The default reading period is 7 days, with a maximum of 90 days and up to 200 meetings in one response; if there are more meetings, Iskra clearly notes that it showed only some of them and suggests narrowing the period.

For Google Calendar, you can add a new event through a pre-filled form. Iskra does not read or sync your Google calendar, and it does not reschedule or delete events already created there.

Microsoft 365 and Outlook.com do not support CalDAV; for corporate email and calendar on your own Exchange server (not Microsoft 365/Exchange Online), there is the “Microsoft Exchange: mail and calendar” skill.

The skill cannot connect to a corporate server that is reachable only on the organization’s internal network without permission from your installation’s administrator.

Overview

Error What it means What to do

Fill in the “Calendar connection”

Not configured The connection is not filled in

group using the settings button

For Yandex, re-create the

The password is of the wrong category password specifically in the

Invalid login or password

or was copied with an error “CalDAV for Calendar” category;

check the login

Ask your administrator

Corporate sign-in The server requires SSO,

whether access with

required password access is disabled

an app password is allowed

Enter the server address explicitly;

for an organization domain, this is

The server address could not

Server not found the address of the service itself:

be determined

Connect the

This is a Microsoft There is no CalDAV

“Microsoft Exchange” skill:

Exchange server on Exchange

it covers mail and calendar

Contact your

The address cannot be reached from

Server on an internal network installation administrator;

the platform

retrying will not help

Retry the action later

The server limited the The provider has temporarily

(the skill does not retry

request rate limited access

automatically)

The meeting was changed in

Meeting changed Refresh the list and

another client while you were

concurrently try again

editing it

Overview

The app password is kept in the platform’s secure storage and is not shown to anyone, including the agent; it never ends up in conversations or system logs. Iskra connects only to your provider’s server. You can revoke the app password at any time in your provider account; the skill will then stop connecting until you create and enter a new one.

Overview

The skill connects Iskra to corporate email and calendar on a Microsoft Exchange server: check your inbox and folders, find and read an email, download an attachment, send an email or a reply (with your confirmation), mark as read, move or delete an email; in the calendar: your plan for the day or week, meeting details with attendees, creating a meeting with invitations sent automatically, rescheduling, editing and cancelling. Credentials never leave the platform: Iskra handles all communication with the Exchange server, and the agent never sees the password.

The skill is intended only for Microsoft Exchange, specifically for servers with the EWS (Exchange Web Services) protocol enabled. For other email, Iskra has separate skills:

“Mail”: Mail.ru, Yandex, Gmail with an app password, any server with IMAP;

Gmail: a Google account with OAuth sign-in (no password).

If you are not sure what kind of email you have, Iskra asks before setup: corporate Exchange, Gmail or other email over IMAP. There is no need to guess: the three skills are built for different protocols and cannot stand in for one another.

The skill works on request, one action at a time. Continuous sync of the calendar and tasks with “Plans” is set up separately, in the “Calendars and tasks” section of Settings (see “Calendar and task sync”). The Exchange calendar and task list can be connected there too, with the “Exchange Server” card, using the same EWS address and the same domain password as here (in the connection wizard, the address is entered in full, with https:// ). The skill remains the tool for email, which sync does not cover at all, and for one-off calendar questions.

Overview

The connection uses your regular domain account password: unlike Mail.ru or Yandex, Exchange does not require a separate “app password”.

1. Get your server’s EWS address from your system administrator: the full form is

mail.company.ru — Iskra adds the path itself. There is no automatic address detection: you cannot work out the address without your administrator.

2. In Iskra: Skill catalog → “Microsoft Exchange: mail and calendar” → add to your

personal profile → on the skill card, click “Needs setup” (later, “Details” → the “Settings” tab opens the same fields). In the “Skill settings” window, the “Exchange connection” settings group opens:

EWS address: the value from step 1;

Login: two formats work, DOMAIN\ivanov (usually more reliable for corporate servers) or ivanov@company.ru . If one format does not work, try the other;

Password: your regular domain account password. Like regular keys, the field is write-only: you cannot view the saved value, only replace it.

3. Ask Iskra to check the connection. If there is an error, the diagnosis tells you what

to fix: wrong login/password, wrong server address or server unavailable.

Enter connection details only through the skill settings form. Do not paste the password into your chat with the agent.

Overview

Check folders: the list of mailbox folders with name, role (inbox, sent, trash, spam, drafts), and the number of unread and total emails.

Show emails: from “Inbox” by default, with the option to choose another folder and show only unread ones. Emails come in pages: Iskra shows up to 50 at a time (usually 20) and tells you how many emails the folder has in total. If not all are shown, ask “show more”: Iskra continues from where it stopped, and so on to the end of the folder. Besides regular emails, the list shows meeting invitations, responses to them and cancellation notices: everything you see in that folder in Outlook.

Find an email: server-side search by sender, subject, message text or date range. Results also come in pages.

Read an email: the message text and a numbered list of attachments.

Download an attachment to the chat files (up to 3 MB).

Send an email and reply to an email: both actions show an editable draft in a confirmation form; the email is sent only after you explicitly click “Send” in the form. For a reply, the server itself builds the subject and the thread from the original email.

Attach files to an email: ask Iskra to attach a file that is in this chat (for example, a downloaded attachment or a prepared document). The attachment names are listed in the confirmation form; check them before sending. Up to 10 files, each up to 3 MB, up to 10 MB in total; the file name and type are preserved for the recipient.

Mark emails as read or unread, move them to another folder, delete them (moved to “Trash”, not permanently): Iskra performs all three actions only after you explicitly agree in a chat message.

Calendar

Plan for the day or week: Exchange calendar events for a period (the next 7 days by default); recurring meetings are already expanded into separate events.

Meeting details: time, place, organizer, attendees, description.

Creating a meeting: with a place, attendees and a description; a confirmation form is shown before the meeting is created, and invitations go to the attendees automatically. Recurring meetings cannot be created yet.

Rescheduling and editing a meeting: time, title, place, description; attendees receive an update.

Cancelling a meeting: it is moved to “Deleted Items”, and attendees receive a meeting cancellation.

Overview

When Iskra shows you the text of an email, in a list or on direct request, the email stays unread on the server, as if you had never opened it in your email client. The same applies to viewing attachments. An email can be marked as read only by a separate explicit action, and Iskra asks for your consent first.

If an email requests a read receipt (the sender turned on the “read receipt” option), marking it as read through the skill may send that receipt to the sender, exactly as if you had opened the email in Outlook and agreed to send the receipt. The skill does not suppress this and does not ask separately, so keep this in mind before agreeing to mark such an email as read.

Overview

All three skills do the same job, working with email from chat, but over different protocols, and the settings of one do not work for another:

“Microsoft Exchange: mail and calendar” (this skill): only for Microsoft Exchange with EWS enabled; the connection uses the EWS address, a login and your regular domain password.

“Mail”: Mail.ru, Yandex, Gmail with an app password, any server with IMAP; it will not work for Exchange without IMAP enabled.

Gmail: Google accounts only, with OAuth sign-in and no password entry.

If you have corporate email and do not know whether it runs on Exchange or regular IMAP, check with your system administrator.

Overview

It works only with Exchange email reachable at the address your administrator gave you: the skill cannot see the organization’s internal network (VPN, private address).

Microsoft 365 / Exchange Online and sign-in with a Microsoft account are not supported in this version.

Automatic server address detection is not supported: always get the EWS address from your administrator.

The skill does not show Exchange contacts. Responding to other people’s invitations (accept/decline) and viewing colleagues’ availability are not supported yet.

You can attach only files from the current chat to an email: the skill will not download a web link or a file from your computer by itself, so attach the file to the chat first.

The default calendar reading period is the next 7 days, with a maximum of 62 days and up to 50 events at a time. Recurring meetings cannot be created yet, only read; they arrive already expanded into separate events.

Overview

Error What it means What to do

Fill in the

Not configured The connection is not filled in “Exchange connection” group

using the settings button

Check the password;

The login, password or login try another login

Invalid login or password

format is incorrect format ( DOMAIN\user or

The EWS address is wrong or Check the EWS address with

Server unavailable

the server is not responding your administrator

Error What it means What to do

The specified folder does not exist

Folder not found Check against the folder list

on the server

The email was deleted or Refresh the email list and

Email not found

its ID is out of date try again

Attachment too large The attachment exceeds the 3 MB limit Download the attachment another way

The server rejected

Check the recipient

Sending failed the recipient, sender

address and try again

or message body

The event was changed or

Meeting not found Refresh the event list

deleted

Overview

The password is kept in the platform’s secure storage and is not shown to anyone, including the agent: it never ends up in conversations or system logs. Email text is different: the “read email” command exists precisely to bring that text into your conversation with the agent; that is the point of the skill. Even so, email text is not written to system logs. Iskra connects only to your provider’s server at the address you specified in the settings. You can change the password at any time through your organization’s IT department; the skill will then stop connecting until you enter the new one.

Overview

The Reputation.ru integration helps you search Russian court cases, read the court rulings it finds and order an in-depth analysis of case law right from an Iskra chat.

Overview

Iskra may offer two versions with the same capabilities:

Version Where the key comes from Who pays for Reputation.ru requests

Reputation.ru ( reputation ) A corporate key configured by the platform administrator Iskra converts the cost of the external request into MT and deducts it from the profile

Reputation.ru —

You enter your own API Reputation.ru charges the cost to your

personal

key in the secure account; Iskra does not add the provider’s

key

skill form cost in MT

The usual MT consumption for the conversation with Iskra itself applies in both versions. With BYOK, only the additional cost of the Reputation.ru request is not charged.

The system version is available automatically if the administrator has enabled it and configured a corporate key. BYOK is in the “Data” category of the skill catalog. If

the marketplace is enabled, install it; if the marketplace is turned off on your installation, enable BYOK mode in the skill catalog. Then open the card and set your personal API key through the secure form. Do not send the key as a chat message.

Even when BYOK is installed for an organization, each member sets their own personal profile key: this skill has no shared organization key. An installed or explicitly enabled BYOK replaces the system version for the users who have access to it. If the personal key is missing or has stopped working, Iskra will not silently switch to corporate access. When you try a request without a key, Iskra opens the secure form and stops until the key is set up. If the form does not appear, open the “Reputation.ru — personal key” card manually. No external request is sent at that point. To go back to corporate access, remove the BYOK installation or, if the marketplace is turned off on your installation, disable BYOK in the catalog.

Overview

Tool What it does

Searches arbitrazh (commercial) court cases with a semantic query

search_general_court_cases Searches cases of courts of general jurisdiction

Searches rulings and reviews of the Supreme Court of the Russian Federation

Starts an in-depth analysis of case law

Checks the status of a previously started analysis

Gets the finished Markdown report and case IDs

Gets the text of an arbitrazh court document from a search result

Gets the text of a general jurisdiction court document from a search result

For example, you can ask: “Find arbitrazh court cases on early termination of a lease”, “Show Supreme Court rulings on subsidiary liability” or “Run an in-depth case law analysis on this issue”.

A search returns no more than 10 compact results per request. An empty result also counts as a completed Reputation.ru request and may be charged. To read a document, use only an ID that the same version of the skill previously showed you in search results: an arbitrary ID or someone else’s ID is rejected before any external request is made.

If the outcome of any paid operation remains unknown after it was sent, the skill does not retry the external request automatically: a retry could lead to double payment.

In the system version, additional MT are charged only after Reputation.ru confirms success. An error response from the provider and an uncertain network outcome are not billed in Iskra. If the provider has already returned a result but the MT charge was not confirmed, the skill keeps the available result or request_id , marks it as unpaid and blocks automatic retries.

Overview

An in-depth analysis is started separately from a regular search and always requires confirmation:

the system version shows the provider’s price and the calculated charge in MT; the value shown is the upper limit of the charge;

BYOK warns that the paid request will be made from your Reputation.ru account, with no fixed price promised.

After confirmation, Iskra sends the request once and returns a request_id . The status can be checked no more than once every 10 seconds, and the finished result is retrieved via deep_result . You can check only an analysis that you started earlier through the same version of the skill.

If the connection drops after sending and Iskra cannot determine the outcome, the paid request is not retried automatically. First check the status using the request_id you already have, or check the charges in your Reputation.ru account; start a new run only deliberately.

Overview

Before an external request, Iskra fetches the current Reputation.ru rates, checks the available limit and calculates the charge from the provider’s price, the platform coefficient and the MT price of your plan. If the rates cannot be reliably verified or the provider’s base cost is above the cap set by the administrator, the request is not sent. The cap is checked before the platform coefficient is applied. Checking the status and getting the result of an in-depth analysis incur no additional provider cost as long as the current Reputation.ru rates confirm a zero price for these operations.

With BYOK, Iskra does not fetch your account’s rates and does not charge MT for requests to the provider. Check the balance, quotas and actual price in your Reputation.ru account.

Overview

The text of a search or analysis request is sent to the external Reputation.ru service. Do not include information that is not needed for the search. The API key is taken from secure storage and is not added to the chat text.

The “Antonym” private mode covers Iskra’s LLM requests, but not the skill’s calls to Reputation.ru: the request text goes to the external service even with Antonym turned on. The search uses real full names and INNs (taxpayer IDs): nothing in the request is substituted.

The skill stores only a limited list of IDs needed for subsequent safe calls:

up to 100 unfinished analyses; an unfinished record expires after 24 hours;

up to 100 finished and expired records, kept for no more than 30 days, with records not used for a long time removed;

the last 100 arbitrazh court document IDs and the last 100 general jurisdiction court document IDs.

A long document or report text is shortened to 60,000 bytes without breaking characters. This is a limit on the skill’s response, not a change to the document on the provider’s side.

The provider’s official documentation:

AI case law search;

in-depth AI analysis;

arbitrazh court documents;

courts of general jurisdiction.

Overview

The skill connects Iskra to your mailbox over the standard IMAP and SMTP email protocols: check your inbox and folders, find and read an email, download an attachment, send an email or a reply (with your confirmation and, if needed, with files from the chat attached), mark as read, move or delete an email. Credentials never leave the platform: Iskra handles all communication with the mail server, and the agent never sees the password.

The skill works with Mail.ru, Yandex Mail, Gmail with an app password, and any server with IMAP enabled, including corporate ones. For corporate Microsoft Exchange email where the administrator has turned off IMAP, there is a separate “Microsoft Exchange: mail and calendar” skill; see “How ‘Mail’ differs from Exchange and Gmail” below.

Overview

Mail.ru, Yandex and Gmail always reject sign-in with your main email password: an email client needs a separate app password. In addition, Yandex has a separate switch in Yandex Mail itself that enables IMAP access and app passwords; without it, the server rejects even a correct app password.

1. Enable access and create an app password with your provider:

Yandex. In Yandex Mail, open Settings → Email clients and turn on “From the imap.yandex.ru server via IMAP” and “App passwords and OAuth tokens”. Then, in Yandex ID → Security → App passwords, create a password of the “Mail” type (Yandex Help). Mail.ru. Mail settings → Security → “Passwords for external apps”. IMAP is enabled by default; the login is your full email address,

Gmail. In your Google account, turn on 2-Step Verification (app passwords are not available without it), then go to Security → App passwords. In Google Workspace, the administrator may block this access; in that case, connect the “Gmail” skill, which signs in via OAuth.

Corporate email. Your regular password usually works; your system administrator can tell you whether IMAP is enabled for your mailbox and what the server addresses are.

2. Find out your provider’s server addresses. For popular providers, these are imap.mail.ru and

You do not have to specify a port: by default, 993 is used for IMAP and 465 for SMTP, both over a secure connection; if your server uses different ports, add them after a colon, for example imap.company.ru:143 .

3. In Iskra: Skill catalog → “Mail” → add to your personal profile → on the skill

card, click “Needs setup” (later, “Details” → the “Settings” tab opens the same fields). In the “Skill settings” window, the “Mail connection” settings group opens:

IMAP server and SMTP server: the addresses from step 2;

Login: your full email address;

Password: the app password from step 1, in full and without spaces. Like regular keys, the field is write-only: you cannot view the saved value, only replace it.

4. Ask Iskra to check the connection. Both channels are checked: reading (IMAP) and

sending (SMTP). If sign-in is rejected, Iskra shows the mail server’s response and a checklist for your provider: what to turn on and which password to paste; if the server address is wrong or the server is unavailable, the diagnosis says so too.

Overview

One installed skill connects up to ten mailboxes with equal standing: for example, your personal work email and a shared address for customer inquiries. The mailboxes are equal: there is no “main” mailbox, no roles and no default mailbox.

In the skill settings form, each mailbox is a separate card with a login and an IMAP server; the “Add connection” button is available until there are ten. Each card can be edited or deleted separately: the other mailboxes keep working, and none of them becomes the “main” one in place of the deleted one.

When there is one mailbox, Iskra works with it without asking.

When there are several mailboxes and your request does not make clear which one you mean, Iskra shows the list of connected mailboxes and asks you to choose. It never picks silently: an email must not go out from the wrong address. You can say right away which mailbox to use: “check mail at mail@iskrabot.ru”.

Iskra handles “Check all my mail” mailbox by mailbox and tells you what it found where.

The confirmation form for sending and replying always shows the exact sender address; check it before sending.

Email IDs are tied to their mailbox: an email from one mailbox cannot be read, forwarded or deleted “through” another, even if the emails have the same folder and number. IDs obtained before the second mailbox was connected stop working; just ask Iskra to show the emails again.

Personal and organization connections do not add up. If your organization has shared mailboxes set up and you create a personal connection, only your personal collection applies to you: none of the organization’s mailboxes are used while your personal collection exists. The ten-mailbox limit is also counted per active collection, not in total. The settings form warns you about this before you create your first personal connection.

Enter connection details only through the skill settings form. Do not paste the password into your chat with the agent.

Overview

Check folders: the list of mailbox folders with name, role (inbox, sent, trash, spam, drafts, archive), and the number of unread and total emails.

Show emails: from “Inbox” by default, with the option to choose another folder and show only unread ones. Emails come in pages: Iskra shows up to 50 at a time (usually 20) and tells you how many emails the folder has in total. If not all are shown, ask “show more”: Iskra continues from where it stopped, and so on to the end of the folder.

Find an email: search on the mail server side by sender, subject, message text or date range. Results also come in pages.

Read an email: the message text and a numbered list of attachments.

Download an attachment to the chat files (up to 3 MB).

Send an email and reply to an email: both actions show an editable draft in a confirmation form; the email is sent only after you explicitly click

“Send” in the form. For a reply, Iskra builds the subject and the thread from the original email.

Attach files to an email: ask Iskra to attach a file that is in this chat (for example, a downloaded attachment or a prepared document). The attachment names are listed in the confirmation form; check them before sending. Up to 10 files, each up to 3 MB, up to 10 MB in total; the file name and type are preserved for the recipient.

Mark emails as read or unread, move them to another folder, delete them (moved to “Trash”, not permanently): Iskra performs all three actions only after you explicitly agree in a chat message.

Overview

When Iskra shows you the text of an email — in a list or on direct request — the email stays unread on the server, as if you had never opened it in your email client. The same applies to viewing attachments. You can mark an email as read only with a separate explicit action, and Iskra will ask for your consent first.

Overview

All three skills solve the same task — working with email from the chat — but over different protocols, and the settings of one do not work for another:

“Mail” (this skill): Mail.ru, Yandex, Gmail with an app password, any server that supports IMAP; to connect, you provide the IMAP and SMTP addresses, the login and an app password.

“Microsoft Exchange: mail and calendar”: only for Microsoft Exchange with EWS enabled; to connect, you provide the EWS address, the login and your regular domain password. This skill also handles the Exchange calendar.

Gmail: only a Google account, sign-in via OAuth without entering a password.

If you have corporate email and do not know whether it works over regular IMAP or over Exchange, ask your system administrator.

Overview

The server must be reachable from the internet: the skill cannot see a mail server on your organization’s internal network (VPN, private address).

The connection is always secure; the skill does not establish unencrypted connections.

You can send only from the address of the mailbox the email goes out from; you cannot choose “someone else’s” sender address.

A personal collection of mailboxes fully replaces the organization’s collection: connections from the two collections are not combined.

You can attach only files from the current chat to an email: the skill will not download a web link or a file on your computer by itself, so attach the file to the chat first.

This skill has no calendar or contacts — only email.

Iskra shows very long emails with the text truncated.

Overview

Error What it means What to do

Fill in the group

Not configured The connection is not filled in “Mail connection”

via the settings button

Error What it means What to do

Checklist: the login is the full email address; Yandex: in Mail, enable “From the imap.yandex.ru server via IMAP” and “App passwords and OAuth

The server rejected the login and tokens”; use an app

Sign-in rejected (wrong

password; in its reply, Iskra password of the “Mail” type;

login or password)

quotes the server’s text Mail.ru: a password for

external applications; Gmail: an app password with two-step verification turned on; the password is pasted in full, without spaces

Check the server addresses,

The IMAP or SMTP address

and if necessary

Server unavailable is wrong, or the server is not

confirm them with the provider or

responding

administrator

The specified folder does not exist on

Folder not found Check against the list of folders

the server

The email was deleted or

Refresh the email list and

Email not found the folder contents

try again

changed

The file is over 3 MB, or Send the file in a separate

Attachment too

all the email’s attachments together email or by other

large

exceed 10 MB means

File not found among chat files There is no file with this name in this chat Attach the file to the chat and give its exact name

The server rejected

Check the recipient’s

Sending failed the recipient, the sender

address and try again

or the email body

Several mailboxes connected The request does not make clear which mailbox is meant Name the mailbox you need from the list shown

The email ID

Show the emails of the right

Email from another mailbox was issued by a different

mailbox again

connection

The email reference was issued

ID outdated before a second Show the emails again

mailbox was connected

Overview

The password is kept in the platform’s secure storage and is not shown to anyone, including the agent: it never ends up in the conversation or in service logs. Email text is different: the “read email” command exists precisely to bring that text into the conversation with the agent — that is the whole point of the skill. Even so, the email text is not written to service logs. Iskra connects only to your provider’s servers at the addresses

you specified in the settings. You can revoke the app password at any time in your email’s security settings; the skill will then stop connecting until you enter a new one.

Overview

Iskra is an AI platform. “Iskra” is the official product name; Anthill and the model name are technical labels. Chat with the agent: attach your materials and ask it to compare sales for two months.

What Iskra can do

Documents and analytics: read PDF/DOCX/XLSX, compare contracts, calculate metrics, create and edit DOCX/XLSX/PPTX, work with PDF. DOCX can be exported to PDF; PDF → DOCX is not supported.

Texts and presentations: emails, instructions and slides; specify the format and style.

Web search: up-to-date information and source checking; name the topic and the period.

Visualizations: diagrams, charts, graphs and interactive dashboards — for example, sales from a spreadsheet with filters.

Images and video: generation and editing are available when the system skill is configured for your installation. Describe the image.

Recordings: where the platform tool is enabled, Iskra transcribes audio and video; a long recording may be processed asynchronously.

Connections: amoCRM, Bitrix24, 1C, email, calendars, Metrica, Webmaster, WebDAV drive, GitHub. Set up the skill with the permissions it needs. Telegram/MAX is enabled by the operator; connect a bot with chat permissions.

Plans: one-time and recurring runs without your involvement, where the “Iskra run” action is available in “Plans”. Set the time and the task, for example a report every Monday; the result stays in the chat and you get a notification. “Auto” mode picks the skills for a request; the schedule is set in “Plans”.

Code and your own skills: explaining code, calculations, repeating procedures. For analysis, documents and multi-step processes, start with a text skill: describe the steps. Regular runs do not require a program skill.

What is possible depends on the installation, plan, settings, tools and permissions. Execution confirms the result; an unknown detail remains unconfirmed.

Materials and working context

To open an arbitrary file from your device, attach it or connect a Memory folder. A file path alone does not give access.

Large materials: you do not need to split a document into parts in advance — Iskra reads it in parts; exact limits depend on the file type and settings.

Subagents: Iskra can delegate independent parts of a single task, as well as checking the result, to them.

A folder as a knowledge base: a connected Memory folder is used without any separate RAG setup.

Long chat: in a personal chat, Iskra itself compacts early turns into a summary when the context runs short; to do it manually, use “Compact” or /compact.

What Iskra knows about you

Iskra takes the available sources into account; chats are not carried over into other chats on their own.

Saved at your request: facts in the “Memory” skill (when it is enabled; long-term — in all your chats, “for this chat” — only here) and /pin until the end of the chat.

Provided from settings and working context: profile fields, learning profile, project instructions, connected folders. In a shared project chat, “About”, instructions and the learning profile are not used.

Current chat: messages and attachments, including “remember” without saving to the skill.

Where to look in the help

Choose a document by topic and the section you need in its table of contents.

Getting started, mobile app.

Chat: “Public link to a chat”, “Chat files”; how the agent works.

Skills: “Creating a skill from chat”, “Where to keep skill materials”, “How to set up a review of the result”; installation and publishing.

Memory and Folder sync.

API keys: “How to download a created file via the API”.

Plans, Calendar, calendar sync.

amoCRM, Bitrix24, 1C.

Mail/Gmail, Exchange.

Metrica, Webmaster, DataNewton, WebDAV drive.

Projects, organizations.

Settings: “Interface language”, “Response language”; promo codes.

Learning, Reputation, apps.

Security, deployment environments, GPU, developer and licenses.

Errors and support.

Interface and response languages are separate settings; the list of interface languages does not limit the languages you can communicate in. Whether the mobile app is compatible with an installation is confirmed by the administrator.

Privacy and current terms

The data route is determined by the installation and its configuration. On-premises deployment by itself does not guarantee that data stays within the perimeter: external models and integrations are possible. The configuration of the specific installation needs to be checked with its administrator. The “Antonym” private mode, chat storage, administrative access and logs are covered in the security section. Contact for on-premises and partnership inquiries: mail@iskrabot.ru.

Your current plan and limits: Settings → “Subscription” (/settings/billing); pricing: /pricing, if billing is enabled. Current prices and your active subscription come from live billing, not from this overview. Help.

Overview

This section answers questions about how Iskra handles your data: the private processing mode, chat storage, administrative access, analytics and logs.

How Antonym works: the private processing mode

“Antonym” is a privacy mode. When it is on, requests are processed by a language model on Iskra’s servers and are not sent to an external model provider. The request text itself does not change: what changes is where it is processed, not its content. That is why the mode suits not only personal data but also trade secrets and documents for official use only — there the value lies in the text itself, and distorting it for the sake of privacy would make no sense.

What the mode covers:

the text of your request;

the agent’s work: tool results, intermediate steps, skill selection, the system prompt, instructions from connected Memory folders;

the model’s text reply.

What stays outside the mode. These five surfaces follow the platform’s regular route, which in the cloud means an external provider:

voice input: the recorded audio goes to the speech recognition service;

text-to-speech for replies: the model’s reply text is sent verbatim to the speech synthesis service. Text-to-speech is currently not enabled in any Iskra interface, so this path is not used in practice; the limitation will apply once it returns;

the “task completed” email: if a scheduled task has email notification turned on, the full result text is sent in the email body through the email delivery service;

submitting a skill for moderation: the manifest and the full source text of the skill are sent;

the summary for the platform administrator: chat titles, first messages and file names across the entire installation.

This also includes skills calling their own external services: Antonym covers Iskra’s LLM requests, not the third-party APIs that a skill calls.

What does not work at all in private mode. Neural generation — images, text-to-speech, audio transcription — is not performed while the mode is on; instead it is openly refused: these tasks are run by an external provider, and the mode cannot quietly send a request there. To use them, turn the mode off.

Mobile push notifications are not on this list: in private mode they arrive without a preview of the reply text — only a message that the reply is ready.

How the mode behaves on failure. Antonym intentionally has no fallback route: if Iskra’s servers are busy, the request is not processed and Iskra will offer to retry. The request is not passed to an external model provider in that case — neither under overload nor in any other failure: the mode has no provision for rerouting to an external model. If Iskra’s servers lack the required capability, Iskra will say so directly, and retrying will not help.

The trade-offs. In private mode a different model answers, so on complex tasks the reply may differ, and the complexity levels are reduced to one. The size of the working context does not change.

Turning the mode on is described in the “Settings” → “Antonym — private mode” section.

Chat storage and administrator access

Chats are product data: they are stored in the installation’s database so that you can return to a conversation from any device. Administrators can view chats only in read-only mode, and every such view is logged — an administrator cannot read your conversations unnoticed.

Secret link, access revocation and already downloaded copies

Board material leaves Iskra only through the board’s external audiences — a named invitation and a secret link. An invitation opens the board after sign-in with the same verified address, so a verified profile stands behind it. A secret link opens without sign-in: the link itself is the key, and you should handle it like a password. The database stores not the secret itself but its fingerprint, so Iskra cannot show an issued link again — a lost link is reissued. The full list of board audiences and how a link is issued are described in the “Boards” section.

Revocation closes access but does not undo what has already been handed out. A revoked or reissued link stops opening immediately, including for a reader who has the page open — at the next check. But a file the reader managed to download or copy stays with them: such a copy cannot be taken back. That is why the decision to share externally is made before the link is issued.

Iskra does not keep named lists of secret link readers: the product does not show who exactly opened the board or whether they read the material. Browser fingerprints and a long-term history of network addresses are not collected for this purpose.

In an organization, a board’s external audiences are opened by a member’s personal permission: it is granted by the organization’s owner and administrator, and it is checked every time the board is opened — revoking the permission closes to readers everything the person has published (see the “Account, profiles and organizations” section). Revoking and reissuing a link, like changes to the organization’s external access policy, are recorded in the organization log; secrets and document contents are not included in the entries.

Product analytics and chat content

Basic product analytics collects metadata: tokens, the model used, latency, errors and billing events — chat content is not included. It also records the client used (Android / iOS app, desktop or browser), the operating system, the mobile app mode and usage time. The platform administrator sees this information per account; one account can use several clients. Signals from an open interface are sent when you return to it and periodically while it is visible.

There is one exception, and it is conditional: if the installation administrator has enabled the weekly activity summary (weekly summary), the first user messages of regular work chats for the week are used to build this summary; temporary chats are not included. In installations where the summary is not enabled, chat content is not used in analytics. Full texts of LLM requests and responses are available only in the debug trace, which is turned off in production.

SIEM, DLP and log export

Iskra has no ready-made SIEM or DLP connectors. Log export (for example, via syslog or Kafka) is added as part of an implementation project — discuss your requirements with the Iskra team: mail@iskrabot.ru.

Overview

This section answers questions about Iskra’s deployment options, certificates and regulatory requirements, as well as licensing for an on-premises installation.

Deployment environments: cloud, dedicated environment and on-premises

Iskra has three deployment environments:

Cloud: a managed installation run by the Iskra team. When answering questions about data location, keep in mind that the main LLM path and auxiliary provider calls depend on the installation’s configuration; individual skills with external providers are enabled explicitly by the administrator.

Dedicated environment: hosted entirely on dedicated infrastructure; the commercial coefficient applied to cloud prices is ×1.3.

On-premises: deployment in your organization’s infrastructure. There are two options: full autonomy on the customer’s GPUs (the current reference configuration for full autonomy is 8 × NVIDIA B200) or use of the customer’s secure LLM API/gateway, in which case no dedicated GPU environment is required.

An on-premises environment by itself does not guarantee that data stays within the organization’s perimeter: data location and the absence of external providers depend on the chosen and verified configuration of the specific installation. The configuration may include external LLMs, an external LLM gateway and provider integrations. When answering questions about data locality, suggest checking the configuration of the specific installation instead of promising a fully closed perimeter based on an on-premises deployment alone. Contact for on-premises and partnership inquiries: mail@iskrabot.ru.

GPU and local LLM for full autonomy

The full-autonomy on-premises option runs language models on the customer’s GPUs: a dedicated GPU environment is required. The current reference for this configuration is a server platform with 8 × NVIDIA B200, and such a site is already in operation. If the customer’s secure LLM API or gateway is used instead of local models, no dedicated GPU environment is required.

Exact sizing — server composition, GPU configuration, model choice and acceptable load in terms of the number of users — is calculated and verified in an implementation project for the specific organization; the help contains no universal requirements table. To request a calculation: mail@iskrabot.ru.

Developer and implementation participants

The product developer and the contractors of a specific implementation are confirmed by the company details and documentation of that installation. The site location, GPU configuration and language model name do not by themselves identify the developer organization. If the available materials do not name it, the name remains unconfirmed. You can get details about your environment from your administrator or from the Iskra team at mail@iskrabot.ru.

On-premises licenses: price and what is included

Iskra on-premises is an annual license for deployment in the customer’s environment:

Simple license: ~$28,000 per year, up to 1,000 users; includes the software, updates and branding. Suitable for a small or medium on-premises environment. Extended license: ~$141,000 per year, unlimited users; the main enterprise option for production use. Maximum license: ~$424,000 per year, unlimited users; for security-heavy and strategic customers with dedicated support. “Source code disclosure” at this level means that the customer’s information security specialists get access to audit the source code under an NDA — without transfer of rights and without publication as open source.

Servers, GPUs, LLMs and tokens, gateway, implementation, integrations, customization, operations and an enhanced SLA are not included in the license and are separate line items in the contract. The monthly equivalent only illustrates the annual contract and is not a separate plan. To request a commercial proposal: mail@iskrabot.ru.

Overview

Iskra’s mobile app (Android and iOS) requests system permissions only when you turn on the related feature. Without any of these permissions, sign-in, reading and text chat work fully — only the dependent feature becomes unavailable.

Where to install the mobile app

The “IskraBot” app is published in the official stores:

Android: RuStore.

iPhone and iPad: App Store.

For a corporate installation, the administrator confirms the app’s compatibility and the server address. The app being available in a store does not by itself confirm that a specific corporate installation is accessible.

Staying signed in

After you sign in with a password or through a connected service, the app remembers your account for up to 30 days. Using Iskra extends this period — after a break of two to three weeks, you can come back without signing in again.

If you have already been signed out, sign in again to the same account and choose the same work profile. Being signed out does not by itself delete saved chats and documents.

Full version of Iskra

The app opens the full version of Iskra: projects, Memory, skills and other sections work the same as on the web. The lightweight Lite mode is temporarily disabled, so you do not need to choose a mode when signing in.

Voice and private mode. The “Antonym” private mode changes where the request itself is processed, but it does not extend to speech recognition. Recorded audio goes to the recognition service via the platform’s regular route, which in the cloud means an external provider. If the content must not leave even as a recording, type your question instead. The full list of surfaces outside the mode is in the Security and data privacy section.

Tip after a screenshot

If you take a system screenshot of an open chat in the Full version, the app shows a small bubble “Permanent chat link — here” pointing to the “⋮” (three dots) button at the right edge of the header. The link recipient will be able to open the chat, and you can close access. While the bubble is visible, tapping it or the “Share chat” item in the three-dots menu immediately creates a public link if there is none yet, or shows the existing one. In the window that opens, you can copy the address, update the published content or close access. If a reply is still being generated, the window warns that the current reply is not yet part of the published chat.

The tip appears only in a saved chat that you can publish: one that is not archived and not read-only for you. It is not available in a new chat before the first message. Within one Full version session, the tip appears at most once per chat and no more than three times in total; switching profiles resets the counter. After the app restarts, the reminder may appear again. The bubble fades out after eight seconds; you can also dismiss it by tapping outside it or pressing Escape. The tip disappears when you switch to another chat.

The tip works on iPhone and on Android 14 and later. Iskra only learns that a system screenshot was taken; it does not read or analyze the screen content. On Android 14+, the system reports that the app detected a screenshot. Iskra explains this once, the first time you open a suitable chat. Screenshot detection is on only while the tip can still appear: it is off in settings, on the sign-in screen, while the link window is open and after the tip has already been shown. The Android permission is granted at install time without a separate prompt. There is no tip on Android 13 and earlier.

The “App capabilities” step

After the first sign-in, the app shows an optional “App capabilities” step: a list of the features it uses, the current state of each permission and a “Configure” button. You can skip this step — nothing is blocked.

You can return to this screen at any time: in the app, open Settings → Application Settings → App permissions.

Which permissions the app uses

Notifications: Iskra’s reply in a regular chat arrives as a push message, even when the app is closed. Usually the reply itself is visible in the notification; if the chat was in private mode, a short line saying that the preview is not shown appears instead of the text — open the reply in the chat. Replies to programmatic requests made with an API key do not create push notifications. After you sign out of an account, the app stops receiving that account’s notifications. The permission is requested only via the “Configure” button on the permissions screen.

Microphone: voice input for messages and sound when recording video for attachments. The prompt appears the first time you tap the voice input button in a chat: first the app explains why the microphone is needed, then the system asks for permission. Once you agree, recording continues on its own — you do not need to tap again. In the Full version on Android, the microphone is available to pages of the selected Iskra server. The system permission does not give third-party pages access to it, including pages embedded in the open page.

Camera (iOS): the “Take Photo or Video” item when attaching a file. iOS itself shows the system prompt at first use; the prompt text explains the purpose. When you record video, the system will additionally ask for microphone access.

Save to Photos (iOS): saving images from the chat to the gallery. The system also shows the prompt the first time you save.

Files and attachments: files are attached through the system file picker; no separate permission is required.

On Android 9 and earlier, the system will additionally ask once for storage access the first time you download a file to “Downloads”. The permissions screen does not show this permission — manage it in Android settings.

If you denied a permission

A regular denial blocks nothing: only the dependent feature is unavailable. When you try to use the feature again, the app explains what exactly does not work and offers to allow it again (Android) or to open settings (iOS, where the system does not show the prompt a second time).

If a permission is blocked in settings

When the system no longer shows the prompt (“Don’t ask again” on Android, any denial on iOS, or a block set by an administrator), the card shows the “Blocked in settings” state and an “Open settings” button, which takes you straight to the app’s settings:

iOS: Settings → IskraBot → turn on the toggle you need (Notifications, Microphone, Camera or Photos). Android: Settings → Apps → IskraBot → Permissions → select the permission and “Allow” (for notifications: Settings → Apps → IskraBot → Notifications).

When you return to the app, the state updates automatically. You can revoke a granted permission in the same place — the app keeps working, and only the related feature becomes unavailable.

Overview

An app is a separate web service that lives in Iskra: it has its own address, its own versions and its own access level. It opens in a browser like a regular website, and Iskra’s intelligence is available to it from the inside: the app calls the platform on behalf of whoever opened it, and the usage is charged to that person.

An app is neither a skill nor a scenario. A skill extends what the agent can do in a chat, a scenario is a saved workflow with a skill, and an app is a standalone service with its own address.

The “Apps” section appears in the left panel only if apps are enabled on your platform. If the section is missing, ask your organization’s administrator.

While you have fewer than four apps, a short “What apps are” panel is shown above the list: what apps are, how to create an app from a chat, how to build one in an external environment and what an app runs on. The “Hide” button removes the panel, and the “How it works” link in the section header brings it back. The “Learn more” link in the panel opens this help chapter.

How to create an app

Click “Create app” in the section header (when the list is empty, the same button is also in the “What apps are” panel). A window with two fields opens:

Name: what the app is called for people. 1 to 80 characters, any language. The name is shown in the list, on the app card and on the sign-in page.

Apps built by Iskra also show it in the title of the viewer’s browser tab.

Address: the part of the link used to open the app. Lowercase Latin letters, digits and hyphens (not at the edges), up to 63 characters. The address is suggested automatically by transliterating the name, but you can edit it manually.

After you click “Create”, the app appears in the section with the “Draft” status, and you land on its card. A draft is already a real app: the address is reserved for it and it counts toward the limit on the number of apps, but it has no versions yet, so it cannot be opened.

How to fill a draft

The draft card has a “How to fill the app” block. Both paths lead back to this same card:

“Create in chat” opens a chat with the app creation scenario; the name, address and the app itself are already filled into the request template, so Iskra will not ask for them again. It will refine the idea, build the app, test it and publish the first version here, leaving the source code to you. The button is present only if creating apps from chat is enabled on the platform.

“Upload archive” accepts a zip archive built with the iskra-app pack command. The platform takes the version number from app.json; the version becomes current after it passes verification.

A third path is to build the app in an external environment. You will need the @iskra/apps package, the iskra-app init → check → publish commands, the platform address and an API key with the app publishing permission. This permission is not in the default set and is not offered in the key creation window — ask your platform administrator for such a key. For API keys in general, see the “API Keys” chapter.

When the first version passes verification, the draft automatically becomes a regular app: its status changes to “Active”, and an entry saying that the draft became an app appears in the change history.

A draft lives until you delete it: it was created on purpose, and the platform does not remove it. While there are no versions, the app can be neither opened nor stopped — there is nothing to stop.

Name and address

These are two different things, and it is best not to confuse them.

The name is for people only. It has no technical effect: you can change it as often as you like, links do not break and the app does not restart. You edit it directly in the card header with the “Change name” button; the app owner and the organization administrator can change it. The new name appears for viewers the next time they open the page, within a minute at most.

The address is part of the link. It also determines where a person who follows your link ends up. The address is changed by a separate action, and the change has consequences — see “How to change the address” below.

Every address that has ever been taken stays permanently reserved for the app that held it — including after the app is renamed and after it is deleted. An address cannot be released, and no other app will get it. This protects against substitution: a link you gave someone can never one day open someone else’s content.

What the app card shows

Header: the name, below it the address and the full link to the app, the status (“Draft”, “Active”, “Stopped”) and the access level. The “Open app”, “Stop” / “Start” and “Change address” buttons are here too.

“App state”: status, current version, creation and update dates.

“Runtime”: how the running app is doing: its state (“Deploying”, “Running”, “Unhealthy”, “Stopped”, “Failed”), version, number of restarts, first launch and last report.

“Access”: who can open the app (see “How to share an app” below).

“Versions”: the list of published versions, a mark on the current one, a button to activate another version and uploading a new archive. The last ten versions are kept.

“Pending versions”: versions that are being verified right now, and the reason for rejection if verification failed.

“Personal secrets”: your own keys and tokens for this app, if it asks for them. Only you see the values: the platform shows them neither to the app owner nor to other users.

“Limits”: the app’s platform limits. Only the organization administrator can change them.

“Latest app messages”: recent lines from the app log.

“Change history”: who did what with the app: creation, publishing versions, changing the name and address, changing the access level, working with secrets.

How to share an app

Who can open the app is set in the “Access” block:

“Owner only”: only you see and open the app.

“Organization”: the app is available to members of your organization. This level is available only for an app created in an organization.

“Everyone”: the app is available to everyone, including people outside your organization. Switching to this level always requires explicit confirmation in a dialog, and for a regular organization member it also requires approval by the organization administrator. Approval is granted not for the address but for a specific version: to publish the next version externally, it must be approved again. In an organization, this level is subject to the organization’s external access policy, described in the “Account, profiles and organizations” section.

You can share an app with two links:

the Iskra sign-in link of the form …/apps/enter/address: with it, the person first signs in to Iskra (or chooses which profile to continue in) and then gets to the app. This is the standard path: the “Open app” button on the card leads exactly there;

the app’s direct URL, shown in the card header next to the address. It takes the person straight to the app.

If a person has no access, they see the same “App unavailable” page with a “Sign in” button — whether the app does not exist at all, access to it is closed or their sign-in has expired. The page is identical in all these cases on purpose, and the app’s name is not shown on it: otherwise it could be used to check whether someone else’s private app exists and what it is called. The person sees the name only after the access check — on the sign-in page in Iskra.

The exception is a stopped public app: its existence is already revealed by the public address itself, so instead of “unavailable” an honest “temporarily unavailable” is shown.

How to change the address

The “Change address” button in the card header opens a window with a field for the new address and a preview of the future link. The permissions are the same as for stopping and starting: the app owner and the organization administrator.

Before confirming, read the list of consequences — it is shown in the same window:

old links will stop opening. There is no redirect from the old address: everyone you gave the previous link to needs to get the new one;

viewers will sign in again. Open app sessions are ended;

the app will restart. A short interruption is unavoidable; the app’s data is preserved — it is tied to the app itself, not to the address;

the old address may keep responding with an error for about a minute, because the platform remembers it for a while. The new address works immediately; in the cloud, the first opening can take a little longer than usual because a certificate is issued for the new address;

for a public app, an address that has already been shared changes, and the organization administrators receive a notification. A warning appears on the card that the address was changed after the version was approved.

No more than three renames per day per app. The counter resets at the start of the next day in UTC. The limit is not cosmetic: every taken address is permanently reserved for the app, and in the cloud a certificate is also issued for each new address. If the limit is used up, the platform refuses and suggests trying again later.

Other common refusals: the address is already taken by another app, the app’s current address was entered, or the address does not meet the rules (only lowercase Latin letters, digits and hyphens, not at the edges, up to 63 characters).

Who sees what

App owner Organization administrator Organization member

only with access level

Sees the app in

yes yes “Organization” or

the list

“Everyone”

Opens

yes yes depending on the access level

the app

Publishes versions and

yes no no

activates them

Requests the

yes no no

“Everyone” level

Moves the app from the “Everyone” level back yes yes no

App owner Organization administrator Organization member

Changes the name and address, stops, starts and deletes the app, works with app secrets yes yes no

Approves the “Everyone” level and changes limits no yes no

A regular organization member does not see a colleague’s private app at all — to them it looks nonexistent rather than closed. They do not see someone else’s draft either, even if it has “Organization” access selected: while there is no version, there is nothing to show. Two users without an organization never see each other’s apps.

An app created outside an organization has no administrator: there is no one to approve, so the owner turns on the “Everyone” level themselves, and the limits remain the platform defaults.

App or board: which to choose

A board shows a finished file — a report, a spreadsheet, a presentation; an app is a working program with its own address, versions, secrets and log.

A board fits when you need to show a result: the reader opens a tab through the board’s audience and sees the version its owner has chosen. An app is needed where a page has to do something on the server side — accept a submitted form, call an external service, act on behalf of the person who opened it. An HTML document on a board cannot do that; its limitations are described in the “Boards” section.

The organization’s external access policy applies to an app’s public address: what it closes and when it requires new approval is described in the “Account, profiles and organizations” section. It does not affect boards — their external audiences are opened by the member’s personal permission “Allow sharing boards externally”. One does not replace the other: the permission to share boards externally does not open an app’s public level, and an approved public app does not share a board externally.

Overview

The Yandex Metrica skill shows data from your Metrica counters in Iskra’s chat: traffic for a period, traffic sources, goals and conversions. The skill only reads statistics — it does not change Metrica counters, goals or settings.

Overview

The skill is in the “Data” category of the skill catalog. If the marketplace is enabled, install the skill; if the marketplace is turned off in your installation, enable the skill in the catalog. Then set your personal Yandex OAuth token in the skill’s secure form (it opens on its own the first time you ask about Metrica, or from the skill card). Do not send the token as a chat message.

How to get a token:

1. Open oauth.yandex.ru with an account that has access to the counters you need, and

create an app. Under access permissions, select “Yandex.Metrica — getting statistics, reading parameters” (the metrika:read permission); the skill needs no other permissions.

2. Copy the app’s ClientID and open the link

3. Confirm access, and Yandex will show the token. Paste it into the skill form.

The same instructions are shown briefly in the skill form as a hint for the token field — that is what Iskra refers you to when it asks you to set or replace the token.

The token belongs to you: even when the skill is installed for an organization, each member sets their own token and sees only their own counters. The token is kept in secure storage, is not shown again and is sent only to api-metrika.yandex.net. If you revoked the token or it stopped working, Iskra will tell you and open the form again so you can replace it.

Overview

Counters. “Which Metrica counters do I have?” — a list with the name, website and access rights.

Traffic. “How many visitors did the site have in the last 30 days?”, “How did traffic change week by week in August?” — visits, visitors, new visitors, pageviews, bounce rate, page depth and average visit duration; on request, a breakdown by day, week or month.

Traffic sources. “Where do visitors come from?”, “How much traffic comes from search?” — visits and visitors by channel (search, direct traffic, links, social networks, ads), with detail down to the specific site or search engine if needed.

Goals. “Which goals are set up and how many leads were there this month?” — the counter’s goals with the number of goal completions and the conversion rate for the period.

You set the period in words (“yesterday”, “for the last 7 days”, “from August 1 to 15”); if you do not specify one, the last week is used. Data for the current day is still incomplete in Metrica. If Metrica calculated a report based on sampling, Iskra will warn you.

Overview

If your account has one active counter, you do not need to name it. If there are several counters, name the site in your question (“traffic for shop.example.ru this week”) or choose a counter from the list Iskra offers. The skill does not build a summary across all counters at once — ask for a report on each site separately.

Overview

“OAuth token required”: the token is not set — fill in the form that opens.

“Metrica rejected the token”: the token is invalid or revoked — get a new one using the instructions above and replace it in the form.

“No access to this counter”: the Yandex account that issued the token has no access to the counter; ask the counter owner to grant access.

“Request quota exhausted”: Yandex limits the number of API calls per day; try again later.

Overview

The current skill gets traffic sources and goal completions as separate reports. For sources, visits, visitors and bounces are available. Goal completions and conversion are calculated for the whole counter over the period, without a breakdown by source: you cannot get them separately for search or ad traffic. The skill does not link goals to a specific ad campaign and does not get its spend or cost per lead. Attributing goals to a specific channel requires a separate export with that segmentation; the site’s goal completions cannot all be attributed to a single channel. To analyze spend and campaigns, you need Yandex Direct — the “Yandex Direct campaign statistics” skill; indexing, search queries, positions and the site’s issues in search are shown by the “Yandex Webmaster” skill.

Revenue from showing other advertisers’ ads on your sites is available through a separate skill, “Yandex Advertising Network revenue”.

Overview

The skill connects Iskra to file storage that works over the WebDAV protocol: Nextcloud, ownCloud, Synology DSM, or a corporate Apache or nginx server with DAV enabled. The agent sees the folder list, searches for files by name, downloads them into the chat and uploads files from the chat back to the server. This skill is separate from Iskra’s own “Drive” (folder sync from your computer) and from the upcoming Yandex Disk skill — in descriptions, the agent calls the storage a “WebDAV server”.

Iskra’s server handles the whole protocol: the skill does not see your password and does not access the network itself. The password is kept in secure storage and used only for requests to the address you specified.

Overview

1. Find out the WebDAV address of your storage:

shown in the Files settings, “WebDAV” section);

Synology DSM: enable the WebDAV Server package; the address looks like

Corporate server: the administrator provides the address.

2. Create an app password if your storage supports them (Nextcloud:

“Settings” → “Security” → “Devices & sessions” → “Create new app password”). The WebDAV app password is separate from the passwords for mail and calendar.

3. In Iskra: skill Showcase → “Your server drive (WebDAV)” → add to your personal

profile → on the skill card, click “Needs setup” (later, the same fields open via “Details” → the “Settings” tab). The “Skill settings” window opens with the “WebDAV server connection” settings group:

Connection name: a short one-word name for the storage without /, for example work or nas. While there is only one connection, it serves only as the card label; when there are two or more connections, it becomes the top-level folder in paths (see “Multiple storage servers”), so the name must be a valid folder name from the start;

Login: the username on the server;

Server address: the full address including the path, https:// is required (Iskra replaces http:// with https://; the connection will not go over an insecure protocol);

App password: a write-only field — you cannot view the saved value, only replace it.

4. Ask the agent to check the connection — it will call check_connection and report

the diagnosis: connected, wrong password, server not found or address blocked.

An organization can set up a shared connection for all employees (skill settings in the organization scope). In that case, before every write the form shows the server address, the connection scope (organization) and the login. For a personal connection, the form shows the profile scope. If these settings changed while the form was open, Iskra writes nothing and asks you to confirm the operation again.

Enter connection details only through the skill settings form. Do not paste the password into your conversation with the agent.

Overview

One installed skill connects up to ten storage servers: for example, a corporate Nextcloud and a home Synology. In the settings form, each connection is a separate card with its own name; the “Add connection” button is available until there are ten. The storage servers are equal: there is no “current” or “main” one, and no tool asks you to choose a connection — the path names it:

One connection: paths are written from the storage root: “Documents/Reports”.

Two or more: the first part of any path is the connection name: “work/Documents/Reports”, “home/Photos”. The storage root (“show what’s on the drive”) shows the connection names as top-level folders. Letter case in the name does not matter: Work/Documents and work/Documents are the same connection. Names in different alphabets or with different whitespace characters that look identical are treated as different.

Search runs within one storage server: “find the report in work”. To find a file across all storage servers, the agent repeats the search for each name from the root.

Copying and moving work within one storage server. To move a file to another one, ask the agent to download it into the chat and upload it to the storage server you need.

Before writing, the form names the connection, the server, the scope and the login. If you change the path in the form to a different connection, Iskra writes nothing and asks you to confirm the operation again.

The connection check (“check the drive connections”) polls each storage server and tells you which collection of connections is in effect — personal or the organization’s.

Identical names. If two connections have the same name (ignoring case: work and Work are the same name), both are unreachable by path until one of them is renamed; the agent names the pair and opens the settings form. A name containing / or invisible characters is not valid as a folder name — such a connection is shown separately in the root and cannot be reached by path.

Transitions. When a second connection is added to a single one, paths change form (the name appears); when only one connection remains again, paths are written without the name — in that case an old path like work/Documents is read as a regular folder named work in the remaining storage server.

Upgrading from an earlier version of the skill. A single connection configured before names were introduced is shown as not configured until a name is entered; you do not need to re-enter the password.

Personal and organization connections do not add up. If your organization has shared storage servers configured and you created a personal connection, only

the personal collection applies to you — the organization’s storage servers are no longer used while the personal collection exists. The limit of ten connections is counted against the collection in effect. The settings form warns you about this before you create your first personal connection.

Overview

List a folder: “show what’s in the Documents folder”: name, size, modification date; up to 100 entries by default, then “show more”.

Search by name: “find a file with the word report”: walks the folder tree up to 3 levels deep (maximum 5), no more than 500 results; there is no search by content.

Download a file: the file is saved to the chat artifacts and opens in the artifacts panel; the limit is 8 MB per file.

Upload a file: the source can only be files from the current chat (attached to a message or created by the agent), up to 3 MB per file; the agent shows a form with the fixed actual path of the source file, the destination folder, an editable name on the server and an “overwrite” checkbox. The upload happens only when you click “Upload file”.

Create a folder: with the “Create folder” button in the form.

Copy, move or rename: renaming is moving with a new name within the same folder; before the operation, the agent shows a form with the source and new paths. The change is made only by the form button; canceling the form or a form timeout copies and moves nothing.

Delete: after you agree in the chat, the agent shows a form with the path; deletion happens only when you click “Delete permanently”, and canceling the form deletes nothing. There is no trash: deletion is permanent, and the file cannot be restored through Iskra.

The skill does not create public links; if you need to share a file, do it in the storage’s own interface.

Overview

Servers with Basic authentication over HTTPS (password or app password) are supported. If the server accepts only Digest, Kerberos or corporate sign-in through a browser, the agent will reply that this sign-in method is not supported — ask the administrator to enable Basic over HTTPS or app passwords.

The server certificate must be trusted: a self-signed certificate causes a secure connection error.

Files up to 8 MB can be downloaded, and files up to 3 MB can be uploaded from the chat (a limit of the chat file channel).

Files and folders are addressed by path: if the server exposes file versions or variants via “?” in the address, such entries are not shown in the list.

If the server returns a reserved character in a name’s address in a form different from the safe address used for the next request, the entry is not shown. For example, + must come as %2B: these two forms cannot be treated as the same file on all WebDAV servers and proxies. Rename such a file in the storage interface or fix the href encoding on the server.

The skill does not follow server redirects. If the server responds to a request by redirecting to another address, the agent will report it — enter in the settings the address that the server considers primary.

An address on an internal network (for example https://cloud.corp.local) is blocked by the security policy by default. It must be added by the administrator of the

Iskra installation to the skill’s list of allowed internal servers — see the “Common errors” section.

Overview

What the agent said Cause What to do

The settings form is not Fill in “WebDAV server

Skill not connected

filled in connection”

Create an app

The server rejected the login or Wrong password, or an

password and paste it

password app password is required

into the form

The server has password sign-in

The server requires sign-in through Check with the storage

disabled for external

a browser (SSO) administrator

applications

Sign-in method not supported The server offers only Digest/Kerberos Enable Basic over HTTPS on the server

Ask the installation

The server is on an internal network,

administrator to add the host to

Internal network address and the Iskra administrator has not

the skill setting

blocked by policy added it to the list

“Internal WebDAV

of allowed servers

servers”

A non-primary storage address Enter in the settings

The server redirects

is specified, or the server the address that the server

the request

returns a browser sign-in page calls primary

Name the storage server in the path:

The first part of the path must There are several connections, but

“work/Documents”

be a connection the path does not start with

(the names are listed in

name the name of one of them

the reply)

Two connections have the same

Connection names Rename one of them

name, ignoring

match in the skill settings form

case

Fix the name in the

Connection name is not The name contains / , .. or

skill settings form

valid as a folder name invisible characters

(one word, no /)

Copying and Download the file into the chat and

The destination path points

moving work upload it to the storage

to another storage server

within one connection server you need

Could not establish a secure connection Untrusted certificate or not HTTPS Check the certificate and the address

File or folder already exists Upload without overwriting, and the folder or name is taken Allow overwriting or choose a different name

Download over 8 MB Send the file by other

File too large

or upload over 3 MB means

The folder is read-only

No permission for this action Check the permissions on the server

or belongs to someone else

Overview

The password does not leave Iskra’s server: the skill receives only the result of an operation, and the password is sent only to the exact server address you specified. Server redirects are not followed at all, so the password and files never go to another address — not even a neighboring one in the same domain.

Connections are made over HTTPS only; every address is checked by IP before connecting. Internal network addresses are opened only according to the installation administrator’s list, and service addresses (link-local, metadata) are always closed.

Neither the password, nor file names, nor content end up in the logs.

The “act without asking again” mode that some skills have is not offered for this skill: writing and deleting always wait for you to click the button in the form. This way, an agent mistake or an instruction hidden in a document you received cannot delete or overwrite anything without you.

Overview

Iskra can keep your “Plans” in sync with an external calendar: events and tasks travel both ways. A meeting created in Yandex Calendar appears in “Plans” — and vice versa, an entry created in Iskra goes to the calendar you set as the target.

Connections live in Settings → “Calendars and tasks”. The second entry point is a labeled button with a calendar icon in “Plans” on the “Calendar” tab, in a separate row above the view switcher. When there are no connections, it is called “Connect a calendar”. While the list has not loaded yet, and when connections exist, it shows the neutral “Calendars and tasks”; when there are connections, a pill with the overall sync status is also shown.

Connections are tied to the active profile: your personal profile and each organization have their own calendars. Switch profiles to see the others.

If the section is missing from settings, sync has been turned off in your installation by the administrator.

Overview

Provider Sign-in method

Yandex, Mail.ru, iCloud login and app password

Google the “Sign in with Google” button

Exchange Server server address, domain login and password

Another CalDAV server server address, login, app password

An app password is not your mail password. Yandex needs a password of the “CalDAV for Calendar” category; a link to the instructions is right in the wizard, next to the password field. For Yandex, Mail.ru and iCloud, the wizard has no server address field at all, and you do not need to look up the address: the card already knows its service’s address — https://caldav.yandex.ru for Yandex, https://calendar.mail.ru for Mail.ru and https://caldav.icloud.com for iCloud.

A mailbox on your organization’s domain connects through the same card. Mail on a company domain is the same service; your address simply is not on the service’s public domain. Choose

your service’s card: Iskra goes straight to that service’s server and, with the card selected, does not query your organization’s domain at all. For Yandex 360, the login is the full email address including the organization’s domain (for example, ivan@corp.ru), and the password is a Yandex app password; you no longer need to work around this with “Another CalDAV server” and type the address by hand. For other services, the login is whatever the service itself expects: for iCloud it is your Apple ID, even if your mail arrives at an address on your own domain.

If you have already connected such a mailbox through “Another CalDAV server”, it keeps working — you do not have to switch to the card. Avoid creating a second connection to the same mailbox: its events would arrive in “Plans” twice.

Everything said about the cards applies to the wizard in “Calendars and tasks”. The “Calendar” skill — a separate setup for one-off requests in chat — has no service cards, and for a mailbox on an organization’s domain you still enter the server address there by hand.

Yandex’s instructions open on the Yandex website, separately from the form: in a new browser tab, or in an external browser in the mobile and desktop apps. When you return to Iskra, continue connecting: the selected provider and the login you entered stay in the form.

Exchange Server is the other way around: it uses your regular domain password and needs no separate “app password”, but the server address is required — Iskra does not look it up on its own. Ask your system administrator for the EWS address and enter it in full, starting with https://. Either the full https://mail.company.ru/EWS/Exchange.asmx or just https://mail.company.ru works — Iskra adds the /EWS/Exchange.asmx path itself.

The wizard does not accept an address without https://. The login works in two forms, DOMAIN\ivanov and ivanov@company.ru; if one does not work, try the other. If the server is internal and Iskra replies that the address leads into an internal network, your installation’s administrator has to allow it.

Microsoft 365 (Outlook.com, Office 365) cannot be connected for sync — neither via CalDAV nor with the Exchange Server card. For one-off work with it, use the “Microsoft Exchange” skill.

Overview

The wizard has three steps.

1. Provider. Choose what you are connecting.

2. Credentials. Enter your login and password (for your own server and for Exchange, also the server address) and

click “Check and continue”. Iskra signs in to the calendar and reads the list of calendars — this takes up to twenty seconds. You can close the window without waiting for the result: the connection is created anyway, and its card appears in the list.

3. Calendars and switches. Select the calendars and task lists you need in

Iskra. By default, the main calendar and the main task list are enabled. A calendar the server does not let Iskra write to (someone else’s calendar shared with you for viewing, a holiday subscription) is marked “read-only” in the list: its entries arrive in Iskra, but nothing goes out from Iskra to it. There are also two switches here:

“Also send the tasks Iskra performs to this calendar” — off by default; the agent’s instructions do not go there, only the tasks themselves;

“Remind me in Iskra about events from this calendar” — on by default; you get each reminder twice: from the calendar and from Iskra. If the calendar’s reminders are enough, turn this switch off here or later in “Configure”.

Closing the wizard at the third step does not cancel anything: the connection has already been created and stays with the default settings.

Google

Google has its own sign-in: it does not accept app passwords.

1. At the “Provider” step, choose Google.

2. Click “Sign in with Google” — Google’s consent page opens.

3. Allow access to your calendar. Iskra returns you to the connections page, and the new

card is highlighted.

Iskra asks Google for access to your calendar and for your email address — the address is used to label the connection card. Iskra does not know or store your Google account password.

Google cannot be connected in the mobile app: Google does not allow signing in inside the app. Connect Google from a computer in the web version — after that, sync works in the app too. Presets with an app password (Yandex, Mail.ru, iCloud) connect in the app as usual.

If there is no Google card in the list of providers, the Google client is not configured on your installation — contact your administrator.

Overview

Providers have different capabilities. The “what is synced” link on a connection card opens the “What is synced” panel — it answers “yes” or “no” for that connection on nine items: “Tasks”, “External attendees and invitations”, “Tags ↔ categories”, “Series exceptions”, “Task start date”, “Meeting link”, “Priority”, “Visibility”, “Busy status”.

The table below is not that panel but a comparison of the providers as a whole: it also covers things the panel does not ask about.

Yandex / Mail.ru /

Google Exchange Server

iCloud / CalDAV

Events yes yes yes

where the calendar where the mailbox

Tasks no

has a to-do list has a task list

yes, invitations

External attendees

yes yes are sent by

and invitations

Exchange itself

Reply to

yes yes yes

invitation

no (tags stay in

Tags ↔ categories yes yes

Iskra)

Meet link

as a line in the

Meeting link yes is read; your own

meeting description

is written

no (they stay in

Reminders yes yes

Iskra)

arrive, but do not

Series exceptions yes yes

go out from Iskra

Task start date yes — yes

Yandex / Mail.ru /

Google Exchange Server

iCloud / CalDAV

Priority yes — yes

Entries are synced within a window around the current date: the distant past and the distant future are not pulled into Iskra, but they do not disappear from the calendar either.

What you need to know about Exchange Server

Reminders stay in Iskra. Exchange does not accept them: a reminder created in Iskra will not appear in Outlook. The reverse problem does not occur — editing a meeting in Outlook does not erase Iskra’s reminder.

Series exceptions go one way. Iskra sees and shows a cancelled or rescheduled occurrence that came from Exchange. But an occurrence cancelled or rescheduled in Iskra stays only in Iskra. And if you change the recurrence rule itself, Exchange rebuilds the series from scratch, and the series’ previous exceptions are lost there.

A task has no time, only a day. An Exchange task never has its own time or time zone: a task whose due date in Iskra includes a time becomes an “all-day” task in Exchange. Iskra determines which calendar day that is from the connection’s time zone, the same way both when it sends the task and when it reads it back. The entry’s own time zone does not affect the task’s due date: two different zones would shift the due date by a day with every sync.

The meeting link travels as a line of text. An Exchange meeting has no separate field for it, so a link added in Iskra is appended to the meeting description as the line “Meeting link: …” and is read back from there. The link of Exchange’s own online meeting is simply read by Iskra.

A recurring entry in an unusual time zone may fail to sync. If there is no way to name the entry’s zone so that Exchange understands it, that recurring entry stays in Iskra and is listed as “Failed to sync”; the rest of the calendar keeps syncing, and a one-off meeting goes through.

An “all-day” entry without a named time zone may shift by a day if the mailbox’s time zone differs from the connection’s time zone. You change the connection’s time zone in “Configure”.

Overview

The “Where new records go” block names the calendar for new events and the list for new tasks. Until a target is chosen for a type, the first enabled matching calendar becomes the target, as the block’s hint says. This happens even when there are several matching calendars: Iskra takes the first one by name instead of asking, so if you connect several calendars, choose the target yourself. The “Keep in Iskra only” option keeps new entries of that type in Iskra.

If you uncheck the calendar that was the target, the target becomes empty, and Iskra does not substitute a neighboring calendar. You asked to remove that calendar, not to move your entries to another one. The “Where new records go” block will show “Keep in Iskra only” for that type: new events (or tasks) will stay in Iskra until you choose a calendar yourself. If they should go to another calendar, choose it in this block right after you close “Configure”. This rule does not apply in the connection wizard: if at the last step you unchecked the suggested calendar and enabled another one, after “Done” the target becomes the first enabled matching calendar of that connection, just as the caption under the list promises.

You can unlink an individual entry from the calendar: in the entry card, choose “stop syncing”. This is not a way to remove an entry: nothing is deleted on either side — the entry stays in Iskra, the object stays in the calendar, and they simply stop exchanging changes.

And the unlink does not last long. The unlink dialog itself says so: syncing stops “until this entry arrives from the calendar again”. It arrives at the first change in the calendar or when the connection re-reads the calendar in full, and Iskra links them again. And if a target is set for this type of entry (“Where new records go”), an unlinked entry counts as new and goes to the target calendar at the next sync. You cannot keep an entry apart from the calendar by unlinking it.

Entries from a read-only calendar are edited in that calendar: time, title, location, reminders and external attendees change there, while in Iskra you can edit such an entry’s project, tags, internal participants, notifications and Iskra’s work. You cannot delete it in Iskra either — it would come back at the next sync. You can remove such entries from “Plans” only for the whole calendar: uncheck it in “Configure”, and Iskra deletes everything that came from it on its side and leaves the calendar itself untouched. The “stop syncing” option does not work for this: the entry stays where it is.

The tags of such an entry remain yours. When the entry first appears in Iskra, the calendar’s categories become its tags, and after that the tags belong to you: syncing no longer overwrites them, even if the entry’s categories change in the calendar. For a regular calendar, tags and categories still sync both ways.

Overview

Sync is two-way, so an entry that leaves “Plans” also leaves the calendar. The object in the external calendar is deleted when its linked entry:

is deleted in Iskra;

is cancelled with the “Cancel” button or declined by you with the “Decline” button — for the calendar, this is the same as deleting it;

no longer falls under the “Also send the tasks Iskra performs to this calendar” switch. Turning the switch off does not just stop sending new items: Iskra’s work that has already gone to that calendar is erased from it.

With Exchange, deleting a meeting with invitees also means cancellation emails: the server itself sends them to real people.

The opposite case is “Disconnect” and unchecking a calendar: they remove entries from Iskra and delete nothing in the external calendar.

Neither “Sync now”, nor pausing and resuming, nor signing in again brings back an entry deleted in Iskra. Iskra remembers the deletion: regular syncing keeps working but does not recreate the deleted entry; otherwise the next sync would immediately undo any deletion. Whether reconnecting the same calendar helps depends on the calendar itself: if the object is no longer there, reconnecting will not bring the entry back; if the object is still visible (for example, sync was paused and the deletion did not reach it), it is intact, and after reconnecting the entry returns to “Plans” once the object arrives in Iskra again.

A cancelled or declined entry is not a deleted one: only the object in the calendar is erased, but such an entry also leaves the regular “Plans” lists, and you cannot bring it back. Restoring the object on the calendar side will not bring back the original entry; if you need the entry again, create it anew.

Iskra’s work tasks themselves are not deleted when the switch is off: turn the switch back on, and the current tasks go again to the calendar currently selected

for tasks in the “Where new records go” block.

You can restore an entry deleted by mistake in two ways. If your calendar can restore deleted items (for example, it has a trash on its website), restore the object there: when it arrives in Iskra again, the entry returns to “Plans”. If there is nothing to restore it from, create the entry anew.

Overview

The status pill on the card shows what is happening:

Syncing — sync is running;

Paused — you paused syncing with the “Pause” button;

Error — sync failed; the reason is spelled out under the label;

Sign in again — access is no longer valid.

Card actions: “Sync now”, “Pause” / “Resume”, “Configure” (label, calendars, switches, time zone) and “Disconnect”. “Sync now” appears only when sync is running or was interrupted by an error: while paused or in the “Sign in again” state, the connection is not synced at all, and the button is not there — first “Resume” or sign in.

“Sign in again” is fixed with one button:

for Google, “Sign in again”: this is the same consent as at the first connection. It happens if you revoked Iskra’s access in your Google account settings or changed your password. The calendars and connection settings are kept;

for the others, “Enter password”: a new app password. Iskra checks it first and only then saves it, so a typo will not erase a working password.

If several entries failed to sync, the card says so and offers a “show” link — it opens these entries in “Plans”.

Overview

Iskra does not re-read the list of calendars on its own, only when you click “Refresh list” in “Configure”, connect the calendar again or sign in again. If by then the server no longer has the calendar (it was deleted, renamed or your access was revoked), Iskra unchecks it automatically, and everything that came from it stays in “Plans” where it was. Entries are not deleted: a calendar missing from a third-party server’s response is not your action, and Iskra will not erase your events and tasks because of it.

What you will see: the calendar is still in the “Configure” list, but unchecked, and syncing for it has stopped. If it was the target for new entries, the “Where new records go” block says so directly: “The target is ‘…’, but that calendar is turned off: new records won’t go there — choose another”.

What to do: if the calendar is back, check it again, and syncing continues from where it left off. If it is not back, choose another target for new entries. Entries that came from the missing calendar stay in “Plans”: to remove them from Iskra, disconnect the whole connection.

Overview

“Disconnect” deletes from Iskra the events and tasks that came from this calendar, together with their reminders in Pulse. Entries created in Iskra stay but stop syncing. Nothing is deleted in the external calendar.

Overview

The app password and Google access are stored in your installation’s secure storage: they cannot be viewed on the page or through the API, only replaced. Iskra connects only to the address you specified, and only over HTTPS.

Overview

Boards are available if this feature is enabled on your installation. If it has been temporarily turned off, the section, publishing and previously issued links are unavailable; the files and the boards themselves are kept.

A board is a separate page where you show finished files to people outside the chat: a colleague, a manager, a client. A board contains only the files you put on it: the conversation, neighboring folders and chat settings are not visible to the reader.

A board consists of tabs. One tab is one file: an HTML report, a text, a spreadsheet, a presentation. A board has one title, one owner and one audience, meaning the group of people who can open it.

How to publish a file on a board

1. In the chat files panel, open the file menu (the “⋯” button) and choose

“Publish on a board”.

2. In the form, choose “Create a new board” and enter a title, or

“Choose a board” from those you already manage. To search by title, use the “Search boards” field.

3. A new board is created with the “Only me” audience: nobody else can see it until you

change the audience yourself. Even if the file is in a project, the board stays private.

4. The form shows the workspace of the future board: your personal space or

an organization. Only you manage a personal board; a board in an organization is managed by you and the organization administrator. The context is chosen once and cannot be changed later.

5. For each file, choose a version mode: “Follow the file” or

“Pin the version”.

6. Click “Review placement”. The “What readers will get” screen opens:

the board title, the audience, the tabs with their version mode and the “Board tabs: … of …” counter. The tab limit is set for the installation and depends on the board’s audience.

7. Click “Place”. You will see “The placement is done” and the number of new tabs.

The confirmation is valid for a limited time, and the form shows until when. If it has expired, prepare the placement again: earlier consent does not carry over to changed material. If the board has changed in the meantime, the form tells you and shows the updated placement.

A file takes up one tab on a board. Publishing the same file to the same board again does not create a second tab: the “What readers will get” screen marks such a file with the note “The file is already placed on this board: no second tab will appear”.

Next to the file in the files panel and in the preview, a placement mark appears: “Placed on a board” or “Placed on … boards”, as well as “Pinned version”, “Showing the version from …” and “Source changed …” when the source has moved ahead of the shown version. “Readers do not see the changes yet” means they still have access to the previous version. If some of the current readers do not receive the material at all, the mark says so separately and offers to update the shown version in the “Access” window.

The preparation status is shown alongside: “Preparing to publish”, “Tab preparation required”, “Board limit exceeded”, “Placement stopped” or “Source deleted”. For example, when the limit is exceeded, the previous ready version may remain available: the reason for the delay by itself does not mean that readers receive nothing.

Which version of a file board readers see

Each tab has its own version mode. For the “Only me”, “Organization” and “Group” audiences, it determines what the reader opens:

Mode What the reader sees

The latest saved and

“Follow the file”

display-ready version of the file

The version you pinned, as long as

“Pin the version”

the pin is in place

For the external audiences, “Named recipients” and “Link”, the shown version of the whole board applies: a ready snapshot of the tabs as of the moment external access was opened. This also applies to personal boards. Later saves of the sources and newly added tabs do not change this snapshot on their own, even in “Follow the file” mode.

When the changes are ready, the owner opens “Access” and clicks “Update the shown version”. The secret link and invitations are kept, and readers get access to the new snapshot of the board. If a tab is not ready for display yet, the window says so: finish the preparation and repeat the update. For a pinned tab, the snapshot still takes the pinned version.

Iskra also saves intermediate edits: the end of a reply does not yet mean the document is ready. For an internal audience, pinning lets you hold on to a finished result while you keep working on the source. For an external audience, publish an updated snapshot once your edits are complete.

You change the mode in the board list: “Pin the version” pins the version readers are getting now, and “Unpin the version” returns to following the file. For an external audience, update the shown version after unpinning so that readers get the changes.

An open page does not swap the document while the reader is reading it. While the browser tab is visible, the page checks the board about every half minute; when a new revision appears, the reader sees “The board has changed” and the “Show the newer revision” button. Downloads and copies also deliver the version shown on screen, not a source that has suddenly been updated.

Who sees a board: five audiences

Audience Who opens the board

The board owner. Every new board

“Only me”

is created this way

Current members of the organization

“Organization”

that owns the board

Members of the selected group in the same

“Group”

organization

Audience Who opens the board

People invited by email address: the board opens after they sign in or

“Named recipients”

sign up to Iskra with that verified address

Anyone who has the secret link; it

“Link”

opens without signing in

A board has a single audience: all its tabs are seen by the same group of people. If materials are meant for different recipients, create separate boards. The “Organization” and “Group” audiences are available for an organization board; a personal board offers “Only me”, “Named recipients” and “Link”.

The audience is changed by a person, with the board’s “Access” button. Choosing a mode does not change access by itself; a separate action applies it. An invitation is sent with the “Invite” button, and a secret link is issued with the “Issue link” button.

An invitation has a deadline by which it must be accepted; revoking an invitation is final, and access can be restored only with a new invitation. A board has one active secret link: reissuing it closes the previous one immediately, and the board no longer opens with it. The link secret is shown only once, so copy it right away; a ready-made address is provided for the recipient.

A secret link does not expire: it works until you revoke or reissue it, as the access window states. There is one exception: an organization board’s link gets an expiration capped by a limit that the people who operate your platform have set, not the organization and not you; in that case, the access window shows until when the link is valid.

The right to read a file is not the right to distribute it. If the file’s source prohibits distribution, the board will not show the file and will say so.

External display of an organization board (“Named recipients” and “Link”) is unlocked by a member’s personal permission, “Allow sharing boards externally”. It is granted by the organization owner and administrator in the employee list; the owner and administrator themselves have this permission by role. Until the permission is granted, the access window says exactly that: sharing boards externally is allowed by the organization owner or administrator, so ask them for this permission. A personal board does not require the permission; its external audiences work right away.

The permission is checked every time the board is opened and applies to the board’s current owner. So revoking the permission immediately closes all of that person’s external boards to readers, while the boards of a colleague who has the permission keep working. Transferring a board to someone without the permission closes it in the same way; transferring it back reopens it.

Revoking access closes the board but does not take back a file that was already downloaded: a copy the recipient already has cannot be recalled.

Where to find your boards

Open “Settings” → “Public links” → the “Boards” tab. It lists all the boards you manage, including private ones: “Only me”, organization boards and group boards. You can search by title and filter by audience and by state, active or deleted.

A board card shows the audience, the owner, the update date, the number of tabs and a “link issued” mark. The “Show board contents” button expands the tabs: here you rename the board and its tabs, change their order with the “Up” and “Down” buttons, pin a version and remove placements.

“Remove the placement” closes the tab’s link, while the file itself stays with you; putting the file back on the board creates a new tab. “Delete the board” closes its links and delivery immediately; the sources stay where they are. A deleted board can be found with the “Deleted boards” filter and restored: it comes back private, and the old secret links do not come back to life.

Boards linked to a conversation are visible right in the chat, in the “Boards of this chat” block of the request panel.

What a board reader sees

The header shows the board title and its tabs; the material shows the owner, the author’s version mode and the time when the shown version reached the board. A tab name may differ from the file name.

HTML opens as a document, a text file as text, and other formats as a card with the “Download file” button. Tabs switch without reloading the page, the browser’s back and forward buttons work as expected, and with many tabs there is an “All tabs” list. Only the open tab runs: neighboring documents are not executed in the background.

On the permanent address, the owner sees the “Manage the board” button and can go to management right from the page. A board opened via a secret link has the caption “Made in Iskra” at the bottom.

If the board is open but has no ready tabs yet, the reader sees “Nothing to show yet”.

Downloading a file and copying it for yourself

“Download file” delivers exactly the shown version of the tab.

“Copy to my chat” creates your own copy of the shown version: the file appears in your files, and the “Open the copy” link leads to the folder where it was saved. From then on it is your file: you can edit it and discuss it with Iskra, and the author’s source does not change.

One tab is copied. Neighboring tabs, the author’s conversation, invitations and the board link are not included in the copy, and neither is the place in the document currently open on screen. The copy takes up space in your storage: if there is not enough space, Iskra asks you to free some up and try again, and for a file that is too large it offers to download it instead of copying.

A board opened via a secret link can be read without signing in, but the copy is saved to your files, so it requires signing in. The “Sign in and copy” button remembers the tab and version, and after you sign in the copy is created automatically.

A link to a tab and to a place in a document

A board’s permanent address looks like /boards/<board>/tabs/<tab>, and the secret link address looks like /b/<token>/tabs/<tab>. The address contains only identifiers, so renaming the board, a tab or the file does not change it. A board address without a tab opens the first available tab.

The “Copy the link” button gives you the tab address. If the document itself reports where you are, such as which slide or which filter, “Copy the link with this place in the document” appears: a #state=… part is added to the address. The caption shows exactly which address will be copied: the filter may contain a value you would rather not send. The link recipient still goes through the regular board access check.

An invalid, revoked or expired link does not reveal the board title. If the link pointed to a tab that no longer exists, the page offers to open the whole board.

What an HTML document on a board can and cannot do

Inside the document, its own markup, styles and scripts work: filters, tabs, slides, charts and calculations on data stored in the file itself.

A document on a board does not go online. Requests to websites and services, form submissions and pop-up windows do not work in it, and it cannot see your profile, your conversation or the board’s other tabs. A request form inside HTML will not submit: build an App for that.

So only a self-contained document reaches readers: everything it shows is stored in the file itself. External addresses include an image from another website, an image at a neighboring path such as logo.png, an image address in styles (background-image: url(…)) and a regular link to a website or an email address. Outbound links found in the document are listed in the “Outbound links” block of the “What readers will get” screen; check it before placing.

A single external address is enough to keep a tab from reaching readers: it stays marked “Preparing to publish” or “Tab preparation required”, and a reader who opens that tab does not get the document. The placement itself still goes through, and the tab is visible in the board’s row of tabs. If the tab already showed a previous version, readers stay on it, and the new one does not reach them.

The fix is a version of the file without external addresses: ask Iskra to embed the images directly in the file and to keep any addresses the reader needs as plain text. Links from the document to other files on the board do not work: a board is not a web directory.

If you want a link to open at a specific place in the document, ask Iskra to record that place in the address: the document reads it from the query string and the URL fragment, and changes it with regular links like #section or through the History API (pushState / replaceState). Then “Copy the link with this place in the document” opens the same slide or filter for the recipient. The app itself does not guess the document’s internal variables.

How to refine a file placed on a board

Material placed on a board remains a regular file, and you can keep refining it. Wherever the file’s placement is shown (for example, in the file card in the “Memory” section), anyone allowed to change the source is offered “Refine in chat”: choose the board and a conversation, and the board is linked to the conversation, while the file’s folder is connected to that conversation’s context. If there is no suitable conversation, choose “New conversation”: the chat is created, and the board is linked to it as soon as you send the first message.

A new board created by publishing from a chat is linked to that conversation automatically and immediately appears in the “Boards of this chat” block. Only a board created in that conversation gets linked: placing a file on an existing board does not link the conversation to it, so link it yourself via “Refine in chat” if you want to see it in the block. A board placed from one chat does not appear in another conversation by itself; a shared folder, a shared source or a shared project does not count as a link.

Then work as usual: the internal audience will see the new ready version of the tab in “Follow the file” mode. For “Named recipients” and “Link”, open “Access” after editing and update the shown version. A pinned tab stays on the previous version until you unpin it.

A reader who has no right to change the source can work with a copy: “Copy to my chat” and then continue with their own file.

What to do if a board does not open

“This version is no longer open” — click “Refresh the board”, and the server shows the current version.

“Sign in to open this board” — the board is closed to an unrecognized reader; sign in with the profile that has access to it.

“This board has reached today’s delivery limit” — the daily delivery limit for the board or its context has been reached; try again later.

“This file only opens as a download” — the format cannot be displayed on the page; download the file.

The link no longer works — the owner reissued or revoked it, or the organization no longer allows the link’s owner to share boards externally. Ask the owner for a new link.

Overview

For owners of websites and apps that display ads from YAN (Yandex Advertising Network): revenue, impressions and clicks for your platforms. Iskra retrieves statistics on what you earn from ad placement: totals for a selected period and daily trends, for the whole account or for a single platform. The skill only reads data.

YAN, Direct, Metrica and Webmaster: which service you need

Data source and capabilities

Your task

of Iskra skills

Yandex Direct skill — campaign

Evaluate the ads you pay for:

statistics. This YAN skill does not

campaigns, spend, cost per lead

receive this data

Yandex Metrica skill. Sources and goals are retrieved as separate reports;

Find out site traffic, traffic sources and goal completions

the conversion of each ad campaign is not calculated in the current skill

Yandex Webmaster skill: SQI, pages in

Check indexing, rankings and errors

search, issues, search queries and their

of your site in Yandex Search

trends

Find out your revenue from ads shown on your

This YAN skill

websites and in your apps

“Yandex Direct ad clicks” in YAN statistics refer to ads on your platforms. This is not a report on the spend of your own ad campaigns.

Connecting YAN and getting an OAuth token

1. Open the YAN dashboard under an account registered in the Yandex Advertising

Network. If you are prompted to register, first finish joining YAN. A regular Yandex ID or a Direct advertiser account does not by itself confirm access to YAN statistics. For a report with metrics, your platforms need to have accumulated statistics.

2. Find “YAN — ad revenue from your own platforms” in the “Data” category

of the skill catalog and install it. If the marketplace is turned off on your installation, enable the skill in the catalog.

3. Open the official statistics API instructions, follow the link for getting a

statistics token and sign in with the account from the first step. Another official way: in the YAN dashboard, click the API icon on the right →

“Get an OAuth token for the statistics API”. A token for the ad block settings API is meant for a different task.

4. Paste only the token value into the rsya_oauth_token field of the skill’s secure

form: without the OAuth prefix, quotes or spaces. The form is available from the skill card; if you request data with an empty or rejected token, Iskra offers to open it from the chat. Enter the token in the form field, not as a chat message.

5. Save the setting and write “Check the YAN connection”, or repeat your report request

with dates. Saving the token does not yet mean access works: it is confirmed by the API’s response to the connection check.

The official token link already contains client_id, the OAuth app identifier from Yandex documentation. It is not your secret. Use the link as is; for this way of connecting YAN, you do not need to create your own app following the Metrica instructions.

If you have a single OAuth token for Metrica, Wordstat or Direct, having those permissions does not by itself confirm access to YAN statistics. You can check the saved token with a connection request; if it is refused, use the official way of getting a YAN statistics token described above.

The token is stored for the current profile and is used only for requests to partner.yandex.ru. When the skill is installed for an organization, each member connects their own token. A revoked token can be replaced in the same form; after saving, repeat your request.

Sample questions

“Show YAN revenue, impressions and clicks for August 1–31, 2026, with daily trends.”

“Which platforms appear in YAN statistics for August 2026?”

“Show statistics for platform 458472 for August 1–7, 2026.”

Specify a particular period and, if you need a single platform, its ID from the list. Iskra gets the available metric names from your account: if the source distinguishes types of clicks, the answer keeps that type, for example “Yandex Direct ad clicks”.

How to read the result

Revenue is shown in the account currency excluding VAT, and days follow the service’s reporting time zone. The answer should include the selected dates and the scope: the whole account or a specific platform. Data for the current day may still be coming in.

The list covers platforms with statistics for the selected period. Platforms without activity may be missing from it; check the full set of platforms in your account in the YAN interface.

If the source does not provide a metric or its value is missing, Iskra reports that it is unavailable. This is different from a numeric zero. An empty report means there is no data for the selected dates and platform. A successful connection confirms access to the API; whether there are statistics for a particular period is checked separately. Iskra reports that there are no rows, keeping the selected period and scope of the report.

YAN connection and report errors

The token is not set, contains extra characters or was rejected — fill it in or update it in the form. After replacing it, repeat your request.

No access (HTTP 403) — open the dashboard under the account the token was issued for and check its permissions; if needed, clarify them with the account owner or YAN support.

Resource not found (HTTP 404) — the saved token was used for the request, but this response by itself does not prove that the token is wrong or that the account is not registered. Open partner.yandex.ru under the token’s account: if a registration page is shown, finish joining; if the dashboard is available, contact YAN support with the HTTP 404 code. Iskra’s internal request_id is meant for Iskra support. Until the cause is clarified, you can leave the saved token unchanged. Automatically repeating the request does not remove the cause.

Request rate exceeded or the source is temporarily unavailable — try again later.

The report was not received in full — first follow the recommendation for the cause of the error: for example, check access or replace a rejected token in the form. If the cause is the export volume, narrow the period and repeat the request. An incomplete export is never presented as the total.

The metric is not available for this report — choose an available metric according to the schema Iskra suggests; for different groups, it may make several requests for the same dates.

The statistics schema is too large — see the full list of metrics in the YAN dashboard. If the schema exceeds the allowed size, the skill refuses to pass it on instead of returning a truncated list.

Overview

The Yandex Webmaster skill shows in Iskra chat how your site is doing in Yandex Search: which sites are verified in your account, the site quality index (SQI), how many pages are in search and how many are excluded, what issues Webmaster has found, which queries the site is shown for, and how impressions, clicks and SQI have changed. The skill only reads data: it does not change sites, permissions, recrawling or the sitemap.

Overview

The skill is in the “Data” category of the skill catalog. If the marketplace is enabled, install it; if the marketplace is turned off on your installation, enable the skill in the catalog. Then set your personal Yandex OAuth token in the skill’s secure form (it opens automatically the first time you ask about Webmaster, or you can open it from the skill card). Enter the token only in the form, never as a chat message.

How to get a token:

1. Open oauth.yandex.ru under the account in which the sites you need are verified, and

create an app: platform “Web services”, Redirect URI https://oauth.yandex.ru/verification_code. Under permissions, select the Yandex

Webmaster permission “Get information about sites” (webmaster:hostinfo); the skill does not need the permission to add sites.

2. Copy the app’s ClientID and open the link https://oauth.yandex.ru/authorize?response_type=token&client_id=<ClientID>.

3. Confirm access, and Yandex shows the token. Paste it into the skill form.

The same instructions are shown briefly in the skill form as a hint for the token field; this is what Iskra refers you to when it asks you to set or replace the token. Unlike Yandex Direct, no API access request is required.

The token belongs to you: even when the skill is installed for an organization, each member sets their own token and sees only the sites of their own account. The token is valid for six months, is kept in secure storage, is not shown again and is sent only to

api.webmaster.yandex.net. If you revoked the token or it stopped working, Iskra tells you so and opens the form again.

Overview

Sites. “What sites do I have in Webmaster?” — a list with the address, verification status and the main mirror, if it is different.

Site summary. “What is the SQI of example.ru?”, “How many pages are in search, and what issues does the site have?” — the quality index, pages in search and excluded pages, the number of issues by level (fatal, critical, possible issues, recommendations) and a list of current issues with the date each appeared.

Search queries. “Which queries is the site shown for?”, “Which queries bring clicks, and at what positions?” — popular queries for a period with impressions, clicks and average impression and click positions, with a filter by device type.

Trends. “How did impressions change over the month?”, “How did the SQI change over the quarter?” — a daily series of impressions, clicks or SQI.

The period is set with calendar dates (“from August 1 to 30, 2026”); if you do not specify one, Webmaster takes the last week, and Iskra shows the actual period boundaries in the answer. If a metric has no data, Iskra says so instead of substituting zero.

Overview

If your account has one verified site, you do not need to name it. If there are several sites, name the address in your question (“SQI of example.ru”, “issues of https://www.example.ru”) or choose a site from the list Iskra offers. Addresses with http://, https:// and www are different sites in Webmaster: a full address with a scheme is matched exactly, and a domain without a scheme is matched by site name. If a main mirror is declared for the site in the account, data is taken from the mirror, and Iskra notes this in the answer. The skill does not build a summary for all sites at once, so ask for a report on each site separately.

Overview

“OAuth token required” — the token is not set: fill in the form that opens.

“Webmaster rejected the token” — the token is invalid or revoked: get a new one using the instructions above and replace it in the form.

“No webmaster:hostinfo permission” — the token was issued by an app without the required access: select the permission in the app settings and get the token again.

“Site rights not verified” — the site is in the account, but your rights to it are not verified in Webmaster; data is available only for verified sites.

“The site has not been loaded or indexed yet” — Webmaster has not collected the data yet; if a different main mirror is declared for the site, Iskra names it.

“Request quota exhausted” — Yandex limits the number of API requests per day; try again later.

Overview

Site traffic, traffic sources and goals are covered by the “Yandex Metrica” skill. Advertiser spend, campaigns and cost per lead belong to Yandex Direct. Revenue from third-party ads shown on your platforms is covered by the “YAN — ad revenue from your own platforms” skill.

Overview

For Yandex Direct advertisers: a list of campaigns with their status and key goals, spend excluding VAT, impressions, clicks, CTR, average cost per click, conversions by Metrica goals and cost per lead for a period. Iskra shows totals for the selected dates, daily trends and a breakdown by campaign, for the whole account or for selected campaigns. The skill only reads statistics: it does not change campaigns, bids or budgets.

Direct, YAN, Metrica and Webmaster: which service you need

Data source and capabilities

Your task

of Iskra skills

Evaluate the ads you pay for:

This Yandex Direct skill

campaigns, spend, cost per lead

The YAN skill — ad revenue from your own

Find out your revenue from ads shown on your

platforms. It does not receive the advertiser’s

websites and in your apps

spend and campaigns

Yandex Metrica skill. It is also a convenient

Find out site traffic, traffic sources

place to get the ID of the goal you count

and goal completions

as a lead

Check indexing, rankings and errors of your site in Yandex Search: Yandex Webmaster. This information is not part of the Direct, YAN and Metrica skills

Conversions in the Direct report are counted by the Yandex Metrica goals linked to campaigns. Traffic and traffic sources for the site as a whole remain in the Metrica skill.

Connecting Direct: your own app and token

Direct grants API access to an app, not to a user, so the token is obtained through your own OAuth app. Iskra does not register an app of its own in Direct.

1. Open oauth.yandex.ru under the advertiser account and create an app. Under

permissions, select “Using the Yandex Direct API” (the direct:api permission); the skill does not need any other permissions.

2. In the Direct interface, submit an API access request for this app in the

“API settings” → “My requests” section. A request is reviewed on business days and takes from an hour to three days, or up to seven in peak periods. Until it is approved, the app’s token is not allowed to access statistics.

3. After approval, open the link https://oauth.yandex.ru/authorize?response_type=token&client_id=<your app ID> under the advertiser account and confirm access, and Yandex shows the token. The app ID is listed on its card at oauth.yandex.ru.

4. Find “Yandex Direct — campaign statistics” in the “Data” category of the skill

catalog and install it. If the marketplace is turned off on your installation, enable the skill in the catalog. In the skill’s secure form, fill in two fields:

“Direct API environment” — leave production: this is the advertiser’s live account. The sandbox value is chosen only to test the app in the Direct sandbox with test campaigns; in the sandbox, a report covers a single campaign.

“OAuth token” — paste only the token value, without the Bearer prefix, quotes or spaces. Enter the token in the form field, not as a chat message. The form is available from the skill card; if you make a request with an empty or rejected token, Iskra offers to open it from the chat.

5. Save the setting and write “Show my campaigns in Direct”. A successful

list of campaigns confirms the connection: the token is accepted and the app is allowed to use the API. Saving the token does not by itself prove access.

If the app’s request has not been approved yet, Direct responds with error 58 “Incomplete registration” (app_not_approved). This is not a token error: you do not need to change or reissue the token; wait for the decision on the request and repeat your query. In this case, Iskra does not open the form but reminds you how the review works.

The token is stored for the current profile and is sent only to the selected Direct API host: api.direct.yandex.com or, in the sandbox, api-sandbox.direct.yandex.com.

When the skill is installed for an organization, each member connects their own token and sees only their own campaigns. A revoked token can be replaced in the same form; after saving, repeat your request. The skill does not support Direct agency accounts with client selection (the Client-Login header): statistics are read for the account that owns the token.

Sample questions

“What campaigns are in my Direct account, and which of them are active?”

“How much did we spend last week, and how many clicks and leads were there?”

“Compare the cost per lead across campaigns for August 2026.”

“Show spend and clicks by day for August 1–31, 2026.”

Specify a calendar period; if you need particular campaigns, name them: Iskra finds them in the campaign list by name and, if several match, asks which ones you mean. For questions about leads, Iskra takes the campaign’s key goal from the campaign list or asks which goal to count as a lead.

How to read the result

Spend is shown excluding VAT in the account currency; Iskra takes the currency from your Direct account settings. Period dates are inclusive; one report covers no more than 366 days, and Iskra splits a longer period into several requests. The answer should include the selected period, the scope (the whole account or specific campaigns), the currency and the note “excluding VAT”.

Conversions are counted only by individual Metrica goals: a “lead” is one specific goal. It is either the campaign’s key goal from the campaign list or the goal you name. When campaigns with different key goals are compared, each campaign’s cost per lead is calculated by its own goal. Iskra does not add up conversions of different goals: a single event on the site can complete several goals, so such a sum does not equal the number of leads. If no goal is specified, Direct’s “all goals” aggregate is shown, and Iskra labels it exactly that way. Conversions are counted using the LC (last click) attribution model.

If the source did not return a metric, Iskra reports that it is unavailable and says in how many report rows the value is missing. This is different from a numeric zero: “unavailable” in the totals means the sum for the period was not obtained, not that there were no leads. An empty report means there are no rows for the selected period and scope; Iskra reports this, keeping the period and scope.

Iskra calculates CTR, average cost per click, cost per lead and conversion rate itself from the report’s totals: impressions, clicks, spend and conversions for the selected goal. If the denominator is zero (no clicks or conversions), the metric is marked unavailable rather than zero.

The Direct report queue

Direct does not build large reports right away; it puts them in a queue. Iskra waits up to several minutes for a report to be ready and then shows the result. If the answer says “the report is still

being generated” (report_pending), repeat the same request in about a minute: Iskra keeps waiting for the same report instead of ordering a new one. The Direct queue holds no more than five reports from one user at a time; Direct keeps finished reports for five hours.

Connection and report errors

The token is not set, contains extra characters or was rejected (not_configured, invalid_token, token_rejected) — fill in or update the token in the secure form; it opens from the chat. After saving, repeat your request.

The app is not allowed to use the API (app_not_approved, error 58 “Incomplete registration”) — the API access request for your app has not been approved yet. You do not need to change the token; wait for the decision on the request in Direct’s “API settings” → “My requests” section.

The account is not connected to Direct (login_not_registered, error 513) — the token was obtained under a Yandex account that has no advertiser account. Get a token under an account that has a Direct advertiser account and replace it in the form.

No permissions (access_denied) — the app or the account is denied access to the requested data. Check the app and account permissions in Direct; if needed, contact the account owner or Yandex support.

Daily units limit exhausted (units_exhausted) — Direct limits API requests with units that are replenished once a day. Repeat the request later.

Request rate exceeded (rate_limited) — try again later.

Report queue is full (queue_full) — the Direct queue already holds five of your reports. Wait until they are ready and repeat the request.

The report is still being generated (report_pending) — repeat the same request in a minute; Iskra continues with the same report.

Direct did not manage to generate the report (report_timeout, HTTP 502) — narrow the period or the number of campaigns and repeat the request.

Direct rejected the request parameters (invalid_request) — Iskra tells you exactly what was not accepted. In the sandbox, a report covers a single campaign: specify one campaign.

The source is temporarily unavailable (source_error) — Direct responded with a server error, or the response could not be received. Try again later; for contacting Yandex support, Iskra provides the request ID if Direct returned one.

The report is too large (report_too_large) — narrow the period or ask for a report on selected campaigns instead of the whole account. A partial total is not shown.

The source response could not be parsed (invalid_response) — Direct returned the report in an unexpected form, and Iskra declined to show it partially. Repeat the request; if the error persists, contact Iskra support.

Overview

The GitHub skill lets you work with a selected repository from chat: read and download files, prepare commits, create and edit issues and pull requests (PRs), leave comments and reviews, view CI results and merge PRs. Iskra acts on your behalf and within the permissions of your GitHub token. Every change requires a separate decision in a confirmation form.

Connecting

1. Open “Skills” → “Catalog”, find GitHub and install it in your personal

profile. If the catalog is turned off in your Iskra, enable GitHub in the list of available

skills. The connection is always personal: even an installation by an organization does not replace each user’s token.

2. Create a personal access token (PAT) in GitHub. For the tasks it supports, use a

fine-grained token: choose the owner, the repositories you need, the expiration and permissions only for your own operations. How to create one is described in GitHub’s documentation.

3. Open the skill settings and enter the PAT in the secure GitHub connection field.

Enter the token only in this form, not in the chat, files or comments. Iskra does not return the token value in answers, created files or regular logs.

4. Optionally, set a default repository in the full owner/repository form and

a default branch. Without them, name the repository and the branch you need in your request; Iskra does not assume main or master.

5. Ask: “Check the GitHub connection and select the my-team/project repository”.

When selecting a new repository, check its full name in the form.

The minimum permissions depend on the action:

Task Fine-grained PAT permissions

Read files and history Contents: Read-only

Create branches, change files and

Contents: Read and write

make commits; merge PRs

Read issues Issues: Read-only

Create and edit issues, comments,

Issues: Read and write

labels and assignees

Read PRs, their changes and reviews Pull requests: Read-only

Create and edit PRs, comment and

Pull requests: Read and write

submit reviews

Read GitHub Actions runs, jobs and

Actions: Read-only

logs

Administration: Read and write and

Create repositories the owner’s permission to create

repositories

Token permissions do not extend your own permissions in GitHub. An organization may require token approval; until the token is approved, there is no access to private repositories. For a classic PAT, an organization with SSO may require separate authorization of the token. For some types of checks, forks and changes to workflow files, GitHub may require additional permissions or another supported token type. If some actions are denied, check the permissions required by the specific GitHub APIs.

The skill works with your PAT and does not charge separate MT for requests to GitHub. Iskra’s regular work in chat is counted under the terms of your plan. Both the cloud version and an on-premises Iskra need network access to GitHub.com and GitHub services for the integration; an isolated installation without such access cannot connect.

Choosing a repository and reading

Iskra works with the repository you selected in the settings or confirmed for this profile. In the result, it shows the full owner/repository name and, when the action involves files, the branch or version. Check them in the form before writing: repositories of different owners can have the same short name.

Sample requests:

“In my-team/project, on the release branch, show the README and the latest commits.”

“Find open issues about export in my-team/project.”

“Show the diff, reviews and check results of PR #42 in my-team/project.”

“Show the failed jobs of the latest CI run and the end of their logs.”

A downloaded file appears among the chat artifacts. The limit for a single file is 3 MiB; Iskra rejects a file that is too large before saving it. A directory listing shows the contents of the requested folder, not necessarily the whole repository tree. Lists and logs may arrive in parts: the has_more or truncated flags mean the output is incomplete. Ask for the next page or narrow the request; the absence of an object in such a response does not prove that the object does not exist.

Changes and confirmation

For a write, Iskra first shows a form with the repository, the target and the exact changes. Review it and confirm the action. Declining, closing the form or not responding does not send the write. Installing the skill and its enablement mode do not replace this confirmation.

Repositories and branches: creating a repository, a fork or a new branch, with the source commit and the destination shown explicitly.

Files: adding, changing, deleting and renaming in a single commit. You can use a file from the current chat. The form shows the paths, the changes, the branch’s base version and the commit message; for binary files, the size and checksum. One operation includes up to 20 changes and up to 3 MiB of new data. If the branch head changed after preparation, the whole commit is rejected: you need to get the fresh version and confirm the changes again.

Issues and PRs: creating, editing the title and description, changing the state, regular comments. For issues, you can add and remove labels and assignees; labels and assignees you do not mention are kept. The assignee’s login is shown in lowercase in the form: GitHub logins are case-insensitive, and Iskra also compares the result in lowercase.

Review: reading discussions and submitting a review of the whole PR. Creating line-by-line comments in a pending review is not available yet.

Merge: a separate confirmation of a specific PR version and merge method. If the PR head has changed, the merge is rejected. Branch protection rules, required checks and GitHub permissions still apply.

Several actions, such as closing an issue and adding a label, are confirmed separately. Iskra does not promise a single transaction across different actions. A successful result contains a link to the object or commit and the observed outcome.

Title or description conflict

Before changing a title/body, Iskra shows the old and new text and then checks the object again. If it has already changed, the write stops before it is sent. After sending, Iskra re-reads the result and reports any discrepancy it finds.

For these fields, GitHub does not offer the same version check on write as for a file commit. A short window remains between the last read and the change: a concurrent edit by someone else may be overwritten, and this cannot always be detected. For text edited jointly, coordinate with your colleagues; if there is a discrepancy, open the link and compare the current text. Iskra does not automatically roll back or overwrite again.

If the outcome is unknown

The connection may drop after GitHub has already performed the action. In that case, Iskra reports “outcome unknown”, saves the operation details and checks GitHub by reading. Ask “check the result of this operation”; if you have a link to the created object, attach it for the check. This does not run the action again.

An incomplete list, no match found or an ambiguous response do not turn an unknown outcome into a failure. If you need a different, independent action, choose a new independent action in the form and confirm its full contents. The previous unknown outcome stays saved; the new operation does not cancel it and may add another object in GitHub.

If Iskra reports that the space for the operation history is exhausted, the new write is not sent. Save the message and contact your administrator; previously unknown outcomes are not reset.

If you connect a different GitHub account while a form is open, the confirmed action is not sent: the form and its contents were built on the previous connection. Repeat your request, and Iskra rebuilds the form for the current account.

Limitations and access errors

The first version works with GitHub.com via PAT. OAuth sign-in, GitHub Enterprise, starting or restarting workflows, deleting repositories and background sync are not among its capabilities. You can read GitHub Actions results.

If access is denied, check the PAT expiration, the selected owner and repositories, the permissions for the action, organization approval and branch protection rules. Show support the error text without the token. If requests are rate-limited, wait for the time GitHub specifies; check an unknown write by reading first. Iskra does not repeat part of a write and does not roll it back automatically.