AI Assistant

How do I open the assistant?

Click the views icon in the top bar and select AI Assistant - next to Calendar, Board and Notes. The same menu appears in the header of each of these windows (the first icon on the left among the buttons), so you can switch views without returning to the top bar. These four windows share the same space: selecting one closes the previous one, and the saved layout (full / docked) applies to all of them.

The floating turquoise circle in the corner of the screen has disappeared - along with the settings for its position and visibility. The assistant now has a single access point shared with the other views.

The window looks and works like Calendar, Board and Notes: use the icon in the header to dock it at the bottom of the screen (it collapses into a pill - click to restore it), switch between the light and dark theme (shared by all four windows), and hide the thread list when you need more room for the conversation. The layout is remembered, so the next time you open it, it will appear just as you left it.

Threads - separate conversations instead of a single stream

On the left-hand side of the window is a list of threads. Each thread is a separate conversation with its own name, history and set of knowledge sources. Your existing history will not disappear - the first time you open the assistant, it is automatically divided into threads based on the context in which each conversation took place.

• New thread - use the “+ New” button above the list. It starts as a general thread (it can only see your tasks); you can add sources yourself using the button above the conversation. To start with the same set of knowledge as the current conversation, select Duplicate with sources from the ⋮ menu. A name is generated automatically from the first question; you can change it by clicking the title above the conversation or using the ⋮ menu.
• Favourites - use the star next to a thread (or the ⋮ menu). Favourites remain at the top of the list.
• Search and filters - the search field searches the names and content of conversations (and displays the number of results). The filters button narrows the list by Workspace, time range, source type, favourites or conversations in which the assistant changed something (created a task, changed a status or set a reminder). When a filter is active, a blue dot appears on the button - even after you close the menu.
• Thread ⋮ menu: rename, add to favourites, Duplicate with sources (a new conversation with the same set of knowledge) and delete. Use the button above the conversation to change the knowledge sources.
• Drag and drop: arrange threads in the list however you like - just drag and move them. Dropping a thread into the Favourites section adds it to your favourites (and vice versa). After the first drag, the list switches to Manual sorting; use the sorting button to select another option.

Knowledge sources - what the assistant can see

Below the conversation title is a Sources button. Context is no longer a single choice (“this task OR that chat”) - it is a set: you can select several tasks, a chat, a note, an entire workspace and the calendar at the same time.

The selection window works like Files - you move deeper one level at a time: workspace → its tasks and chats → thread → its folders, files and notes; plus separate sections for Documents (your private folders) and Mail. The same rule applies at every level: clicking a row selects that location as a source (a tick appears on the right), while the › arrow takes you deeper. The “‹” bar at the top takes you back one level. The search box at the top searches everything at once.

Available source types:

• My tasks - the default; a list of your tasks, as before.
• Calendar - upcoming due dates (7 days by default).
• Workspace - tasks and chats from the entire workspace.
• Task / Chat / Note - a specific item selected from the tree or search results. Selecting a task or chat also includes files uploaded to it - both those in messages and those added in the Files module (shown in the tree as “Workspace / thread name”). You do not need to find them separately; the thread name shows how many files it includes. A long note is read in full - the assistant processes it in sections and fetches more when needed. For you, it remains a single note: you do not need to split or shorten anything, and notes and files are handled in the same way (since 2.2.272).
• Files - “All my Files”, a selected folder (including its subfolders) or an individual file; AI reads the contents of text documents (TXT/CSV/DOCX/XLSX/PDF). You will find workspace folders by opening that workspace, and private folders by opening Documents; use the arrow to open a subfolder, just as in the Files window.
• Mail - your inbox (a list of threads plus the latest message content) or an individual email thread (which you can find using source search). The Mail - inboxes section only appears if an administrator has enabled mail as an AI knowledge source (Settings → AI, disabled by default), and only for your inboxes - no one, including an administrator, can select someone else’s mail. Remember: the content of the selected source is sent to an external AI provider; mail is never added to the knowledge base indexed in the background.

Below the list, you can see a capacity bar showing how much knowledge fits. If you select more than the model can handle, the assistant will tell you exactly what it did not take into account (broad sources are omitted first, while explicitly selected tasks remain). You only see content you have access to - if you lose permission to a source, it will be omitted the next time you ask a question.

Attachments in the assistant panel

The message field has a paperclip (📎) - you can attach files directly to your conversation with the assistant. You can also paste a file or screenshot from the clipboard (Ctrl/Cmd+V) straight into the conversation field - it will be added to the conversation like a paperclip attachment. AI will read the files when responding (supported formats are listed below). Files attached this way are session-specific (they apply to the current conversation).

What you see while waiting for a response

While the assistant is working, a short description of the current stage appears next to the animated dots - for example, “Reading attachments…”, “Reviewing sources: my tasks, mail”, “Translating…”, “Thinking…”, and, if it takes longer, “Writing a response…”. These stages reflect the assistant’s actual workflow and what you selected in the conversation (knowledge sources, attachments and the type of request).

What can the assistant do?

The assistant knows your tasks, chats and workspaces. It can:

• Answer questions about your tasks, due dates and assigned people
• Send emails - for example, “send an email to jan@firma.pl with a summary of task X6”.
• Generate downloadable files - PDF, Word (DOCX), Excel (XLSX) or CSV. For example, “prepare a PDF summarising task X6” or “generate a Word report”. The assistant will retrieve the task content, create the document and provide a download link. When sending an email, the assistant uses your personal SMTP (Settings → Profile → “Send from your email address”). If you have not configured it, the assistant will let you know and ask you to enable your own SMTP configuration - the message will not be sent.
• Access the content of any task or chat you have permission to view - without opening it. Simply refer to it by name, for example, “summarise the Recruitment task” or “what was agreed in the Marketing chat?”. The assistant will then retrieve its full content, messages and attachments (TXT/CSV/DOCX/XLSX/PDF).
• Perform actions in the application: create tasks, open windows, filter the list and change statuses
• Reschedule calendar events - for example, “move all today’s events to tomorrow” or “move Friday’s meetings to the next working day at 12:00”. Before anything is moved, you will see a confirmation card; afterwards, the assistant will list exactly what it changed. Event times remain unchanged (unless you provide a new time), while reminders linked to task due dates are not moved - their timing is calculated from the task’s due date.
• Create calendar events - with start and end times, a workspace, attendees and a colour, for example, “create an event called Offer review tomorrow from 12:45 to 13:30 in Technical, with Kasia Dąbrowska, colour green”. By default, events are created without an email notification - request one separately (“send it by email too”).
• Prepare automations from a description of the rule. A completed draft does not take you away from the conversation: a “Go to Automations” card appears below the response - click it when you want to review and save the draft.
• Help you write task descriptions, emails and notes
• Inform you about overdue tasks and absent colleagues

The assistant can only access tasks and chats that you have permission to view (administrators can access all of them). For very broad questions (“search all tasks…”), it may only retrieve a few of the most relevant items at a time.

Example commands

• “Show my overdue tasks”
• “Create task X with a due date of Friday and assign it to Anna”
• “Change the status of task X to done”
• “Go to the HR workspace”
• “Open the settings window”
• “Show chats only”
• “Sort tasks alphabetically”
• “How do I add a new user?”

Improve text (editing in the editor)

In the toolbar of every editor (task description, task/chat message and message editing), you will find a magic wand with stars icon ✨ - Improve text.

When clicked, AI edits your text naturally directly in the editor: it corrects typos, spelling, grammar and punctuation, and improves the style - writing naturally and clearly (like a thoughtful person, without stuffy formal language). It uses the subject and description of the current task or chat to keep terminology consistent. It does not invent new facts or details, or make the text more formal.

It preserves your layout. Existing formatting remains untouched: bulleted and numbered lists, bold text, paragraphs, emoticons and icons (✅, →), as well as spaces in units (“200 ms”, “5 min”). Improve text only edits the content within the existing structure - it does not turn paragraphs into lists or vice versa. It leaves sections containing a link in the middle of a sentence unchanged to avoid losing the link.

A “Text improved - click to undo” message will appear - clicking it restores the previous version.

The button is only visible if an administrator has configured AI and you have permission to use AI in the relevant context (tasks or chats). The feature only changes the text in the editor - nothing is sent or saved until you send or save the message yourself.

AI Assistant in task and chat threads

The reply field in every task and chat (as well as in Mail) has two labelled AI buttons:

• AI Assistant (purple orb) - switches the editor to private question mode. The editor turns purple, and the bar above it says: “Question for the AI Assistant - only you will see it; it will not be shared with the participants”. The question and answer are visible only to the author - the answer appears in the thread with an ONLY FOR YOU badge. Click the button again to return to regular messages.
• Organise - click to open a menu with four thread reading modes: Original content (disables AI), Polish (the same content, shorter and better organised, in the first person), Summarise (1-2 sentences per message) and Expert (professional wording with an “Expert note” - advice, a risk or a question worth asking). AI processes messages in batches of 8 (starting with the newest); a banner with the mode name appears at the top of the thread, with an “Organise older messages (N)” button below it that loads the next batches. The result applies only for the duration of the session and is not saved anywhere; each processed message includes the note “AI can make mistakes”. Once a mode has been processed, switching is instant - you can move between modes within the same session without waiting.

In question mode, the assistant can see the thread content and attachments, and can also generate a downloadable file (Word/PDF/HTML), e.g. “prepare a Word document summarising this thread”. AI writes the content, while the application creates the finished file and adds a “Download …” button to the thread.

What you can ask for in natural language

There are no special commands - simply ask for what you need in a normal sentence, either in the AI Assistant window or in question mode in a thread. Examples:

A few rules that always apply:

• Sending an email always requires your confirmation - AI never sends emails on its own. Your personal SMTP is used (Settings → Profile → “Send from your email address”).
• Reminders and scheduled actions are handled by the AI Agent - as before (details below). Relative dates and times (“tomorrow at 9:00”, “in a week”) are converted into specific dates.
• Filtering the list of tasks/chats only works in the assistant window (e.g. “show favourite chats”, “high priority, in progress”).

Creating tasks from a thread

In a task or chat thread, use the “AI Assistant” button to enable question mode and ask it to create a new task - just as you would in the side panel:

• “Create a new task with the same brief and a due date of 30 July”
• “Create a task: prepare the quarterly report, due tomorrow”
• “Create a task: Quote for Meblex, due Friday, assign Marek”

The assistant will compile the details and open the “Add Task” window with the fields already completed: the current thread's Workspace, subject, description, due date, priority and assignees. The Task is only created when you click “Save” - before then, you can edit any field or close the window (in which case nothing is created).

A few useful details:

• “with the same brief” copies the current task's description exactly as it is (without summarising it). Without this request, the assistant creates a new description based on what you wrote - it does not copy someone else's brief.
• Dates expressed in words (“30 July”, “Friday”, “in a week”, “tomorrow”) are converted into a specific date.
• The window only opens for you - other thread participants cannot see your conversation with the assistant. Refreshing the page or reopening the thread will not open the window again.

Your conversation with the assistant in a thread is always private - only you can see the question and answer (including a reminder that will appear at the scheduled time). The task created when you click “Save” is a regular task and is visible to participants.

Checklists you can tick off

You can ask AI to create a list with items you can tick off - either in the assistant panel or in question mode in a thread:

• “Create a checklist: call Michał, reply to NBP, issue an invoice”
• “To-do list with checkboxes: …”

AI returns the items as clickable cards with checkboxes. Click an item to tick it off (click it again to undo). The checked status is shared with everyone who has access to the message - when anyone ticks off an item, all other participants will see the same status (it is saved on the server and updated in real time). Each completed item therefore includes a label showing who ticked it off - a mini avatar and first name on the right, with the full name, date and time shown on hover. The label disappears when someone unticks the item. Lists from a private conversation with the AI Assistant in a thread are visible only to you, so their checked status also remains private (there are no labels there - they would always show you anyway). Regular bulleted lists (where you did not ask for items to be ticked off) are displayed as before.

File generation - how it works

When you request a file (in the assistant panel or in question mode in a thread), AI prepares the content, while Task Manager creates the file itself (Word, PDF, Excel/CSV, HTML) and provides a download link. This ensures the file is generated reliably - even if AI “forgets” the format. Word/PDF/HTML files are created directly from the AI response; spreadsheets (Excel/CSV) require tabular data.

When something goes wrong

If AI does not respond, the thread (or panel) will display the specific reason, e.g. timeout (try again or select a faster model under Settings → AI), invalid API key, request limit exceeded or unavailable model. You will not be left without an explanation - you will always be told what to do next.

Reading attachment contents

AI can see attachments added to a task/chat and read their contents - both in question mode in a thread (the “AI Assistant” button) and in the assistant panel (for the currently open task/chat), e.g.:

• “summarise the contents of the abc.docx attachment” (question mode in a thread)
• “summarise the contents of the raport.pdf attachment” (in the assistant panel, with this task open)

Tip: the assistant can access the content and attachments of any available task/chat by its name or from the open view - you do not need to open it. If an attachment was added just a moment ago, the assistant will see it when you ask your next question (the context is retrieved live from the database).

Supported text formats: TXT, CSV, MD, JSON, XML, HTML, DOCX (Word), XLSX/XLS (Excel) and PDF. Images (JPG/PNG/GIF/WebP) are analysed visually if the AI model supports images - including large images (they are automatically resized before being sent to the model, so even large photographs can be described).

Once summaries are enabled, the assistant panel can also answer questions about attachments from MULTIPLE tasks/chats at once (e.g. “summarise the information in attachments from all tasks”) - within your permissions (only tasks and chats you have access to).

Lots of attachments in one thread? In Settings → AI, the administrator can enable “AI attachment summaries”. The application then creates a short summary of each file (text or image), allowing the assistant to handle even dozens of attachments (for example, “summarise all attachments”). Modes: automatic (in the background, requires cron) or manual (the administrator clicks “Process now”). Summaries use one API call per file. Files that have not yet been indexed are read on demand (up to a limit), and if there is too much data, the assistant will ask you to narrow the request or specify a particular file by name.

It reads long documents in sections. When a file cannot be processed in one go (for example, a document dozens of pages long), the assistant receives the first section together with clear information about how much text has been omitted - and retrieves further sections itself when needed. Previously, the text would cut off mid-sentence without any indication, which could lead the assistant to suggest that the file was corrupted or incomplete.

Limitations: older DOC files (binary format) cannot be read. Scanned PDFs (images without a text layer) cannot be read (no OCR). Very large files are truncated (to approximately 50 KB of text), so the AI may only see the beginning of a lengthy document.

Conversation history

The assistant remembers the context of the current thread - each conversation has its own history and is kept separate from the others. The history may be stored in the database (depending on the administrator’s settings) or reset when the window is closed. Threads are private - no one else can see them, including the administrator.

AI settings (administrators only)

Go to Settings → AI tab. Set:

• AI provider: OpenAI, Gemini, Claude or Grok
• The provider’s API key
• AI model - select one from the list retrieved from the provider’s API
• Maximum response length - token limit (1024-16384)

Reminders

The assistant can set your personal reminders (separate from tasks and chats):

• “Remind me today at 18:15 about my meeting with Marek”
• “Remind me in 30 minutes to send the report”
• “Remind me tomorrow at 9 about the invoice”

The assistant will display a confirmation with the date and time - once approved, the reminder is saved. At the specified time, you will receive an in-app notification (toast).

E-mail only on request. Events created by the assistant have no e-mail notification by default - you receive an in-app notification and nothing else. If you also want an e-mail, say so explicitly: “remind me tomorrow at 9 about the invoice and send me an e-mail too”. (The same option appears in the calendar as the “Also send an e-mail notification” checkbox.)

An event with full details. In one sentence, you can provide everything you would set manually in the calendar:

• start and end times - “Offer review event tomorrow from 12:45 to 13:30”
• Workspace - “...in the Techniczne Workspace” (the name is enough)
• participants - “...with Kasia Dąbrowska and Ola Wójcik”; each person receives their own copy of the event (and their own notification)
• marker colour - orange, purple, pink, blue, green, grey

You can only add participants from among the people with whom you share a Workspace. If a first name matches more than one person (for example, just “Piotr”), the assistant will ask for their full name - it will not invite anyone at random.

Important: the reminder is only created after you click “Yes, proceed” in the confirmation window. The assistant’s response alone does not save anything - if the confirmation window did not appear, the reminder was not created. You can check the list of reminders that have actually been saved with the command “show my reminders”.

E-mail requires CRON (an administrator task). In-app toast notifications always work, but sending an e-mail when the application is closed requires a recurring server task:

* * * * * /usr/bin/php /ścieżka/do/aplikacji/ai_agent_runner.php >> /ścieżka/do/aplikacji/logs/ai_agent.log 2>&1

Hosting services with HTTP cron support (cyberfolks, nazwa.pl) can instead request https://domena/ai_agent_runner.php?token=… every minute (the token is in includes/config.php). If the application detects that cron has never run, it will display a warning when you set a reminder - without cron, e-mails will only be sent when someone happens to be using the application.

Reminders linked to a task’s DUE DATE

Instead of providing a date, you can link a reminder to the task’s due date:

• “remind me about this task one day before the due date” (when asking in the task thread)
• “remind me 2 hours before the due date of the task Oferta dla Meblex” (assistant panel)
• “remind me in the morning on the due date” / “...when the task is due today”
• “remind me at the due time” - exactly at the task’s due time

The application calculates the reminder time automatically from the task’s due date (if no time is specified, it uses 9:00). If you change the task’s due date, the reminder moves with it - you do not need to set it again. If the “before the due date” time has already passed after the change but the due date is still in the future, the reminder will be triggered at the next opportunity; if the due date has passed or is removed, the reminder is cancelled.

A task without a due date cannot be monitored in this way - the assistant will explain this and suggest a standard reminder for a specific date and time.

Daily task summary (“due today / overdue”)

In Settings → Profile → Daily task summary, you can enable one notification per day at a chosen time (8:00 by default), summarising how many tasks are due today and how many are overdue, together with their titles. The summary only includes tasks assigned to you (not completed or archived), arrives once a day and is delivered in the same way as reminders

• as an in-app notification, or by e-mail when the application is closed (requires CRON).

Management: “show my reminders” (a list with dates, times and IDs) and “cancel reminder …” (cancel using the ID from the list). If you specify a time that has already passed today, the assistant will ask whether you mean tomorrow; if no time is specified, it uses 9:00. Limit: 50 active reminders.

Verifying AI actions

The application ensures that the AI does not claim to have taken actions that it did not actually perform. If a response contains a confirmation (for example, “I have set the reminder”) but the action was not actually completed, you will see a yellow “Correction” panel below the response explaining what actually happened and how to achieve the intended result (for example, which phrase to use to open the task form or create a real reminder). Sending e-mails always requires you to confirm by clicking “Send”.

Mail as a knowledge source

Mailboxes and e-mail threads can only be selected as knowledge sources if the administrator has enabled this in Settings → AI (“Allow mail to be used as an AI knowledge source”). It is disabled by default because correspondence contains third-party data and is sent to an external AI provider. Once enabled, each user can only see their own mailboxes when selecting sources.

By selecting a folder from Documents, you give the assistant access to both the files and notes it contains (including those in subfolders).

Internet access

If the administrator has granted you the AI Assistant with internet access permission (in the Users window), the assistant can retrieve the content of a specific website - for example, to check whether the domain in a client’s e-mail address exists and what is on the website. Simply ask: “check example.com and tell me what this company does”.

This is not a search engine - the assistant must know the website address (either provided by the user or derived from the domain in an e-mail address). Without the permission, the assistant will clearly state that it does not have internet access.