How does context work in the AI Assistant?
Context in the AI Assistant in Task Manager starts with your tasks. If you need other materials, you choose them as conversation sources. You don't need to explain that you're talking about a quote for Meblex if the conversation already includes the task, a chat about changing requirements and a meeting note. You can simply ask: “What still needs to be agreed?”
Each conversation is a separate thread with its own history and set of sources. In one set, you can combine two tasks, a chat, a note and seven days of calendar entries. A conversation about recruitment won't get mixed up with one about invoices. Threads are private - no one else can see them, including administrators.
The Assistant also knows your tasks, chats and workspaces. In the Pro plan, there's also an indexed company knowledge base: it includes summaries of tasks and conversations, as well as excerpts from original notes and files. The programme updates it once a day when someone uses the application, unless the administrator has disabled automatic updates. No cron is needed for this. For a question of at least 12 characters, the programme can search for matching entries in this base and add them to the response context.
What the Assistant can see
Choose sources using the button above the conversation. The window works like Files: you move from a workspace to its tasks and chats, and then to folders and notes. Clicking a row selects a source, while the › arrow takes you deeper. Selecting a task or chat also includes files attached to it - both from messages and from the Files module. Next to the task or chat name, you can see how many files are included in that thread.
You can point to My Tasks, Calendar (by default the last 30 and next 90 days), the whole workspace, a task, a chat, a note, your own folder with subfolders, or a single file. Only your own files from a folder will reach the model; you can't select another person's folder or a task's or chat's folder as a separate source. Mail is only available if the administrator enables it in Settings → AI. You can only point to your own inbox; not even the administrator can point to someone else's. The assistant reads a long note in portions and fetches further ones when it needs them.
The limit is 30 sources and 60,000 characters. A usage bar appears below the list. If you exceed the budget, the Assistant will list what it left out. General scopes are dropped first; explicitly selected tasks remain. The application checks access to a source when it's saved and again when building the question, because permissions can change.
What do you gain?
You don't have to rewrite an introduction for every question. Without context, you have to remind the Assistant which client you're talking about and where you recorded the agreements. Here, you can ask a short, specific question instead of getting a generic answer about businesses.
Return to a client after two weeks without having to reconstruct the whole conversation. Its history and materials are at hand, including an attachment uploaded on Friday evening. Parallel matters don't get mixed up.
You don't have to wonder whether an answer will include a colleague's private note that they haven't shared with you. Not even an administrator can read it unless it's shared. They can't see your conversations with the Assistant either.
What it looks like at work
A graphic designer manages three clients, each in a separate workspace. In a thread about one of them, they add a task with the brief, a chat with the client's feedback on mock-ups, and their own folder of source files. They ask: “What hasn't been approved yet?” They get an answer based on those three places without opening each one.
A week later, the client sends an attachment with feedback. The designer doesn't copy it into the conversation. The file attached to the task is included in its context, so the designer can ask about the feedback in the file. For larger files, the Assistant may only see an excerpt: previews of files attached to a task have a combined limit of 8,000 characters. For a new matter, they start a thread using the “+ New” button. To keep the same set of materials, they choose “Duplicate with sources” from the ⋮ menu.
Frequently asked questions
Can the Assistant see tasks across the whole company?
No. The Assistant can only access tasks and chats you have permission to view; an administrator can access all of them. A source you lose access to will be skipped the next time you ask a question.
Is task content sent to the AI provider?
Yes. The content of the selected source is sent to the external AI provider chosen by the administrator in Settings → AI: OpenAI, Gemini, Claude or Grok. Email is never sent to the background-indexed knowledge base.
What happens if I select too many sources?
The application will trim the selection and tell you what it left out. Broad scopes, such as an entire workspace, are removed first. Tasks and notes you selected explicitly are kept.
Has my old conversation history disappeared now that threads have been introduced?
No. The first time you open it, your existing history is automatically split into threads based on the context in which the conversations took place.
Could a colleague's private note appear in a response?
No. Access to a note depends on its owner and who they have shared it with, not on workspace membership. Even an administrator in the same workspace cannot see a private note.