What is Company Knowledge in Task Manager?
Company knowledge for the AI Assistant is the material it draws on when answering: tasks, chats, notes, files and folders. Instead of giving generic answers, it draws on what your team has already written down - for example, a task description from six months ago or a note with a procedure. You get the whole company's knowledge base, built in the background, in the Pro plan.
In the Pro plan, this knowledge grows as you work day to day. With activity in Task Manager, it attempts to update the knowledge base with tasks, chats, notes and files once a day. It builds short summaries of arrangements, problems and conclusions from tasks and chats. It doesn't summarise notes and files - it indexes fragments of the originals, starting with the title. For files, it indexes the readable text, not scans or images. This lets the assistant find a specific rate, clause or code within a document.
You can also select sources yourself. The “Sources” button above the conversation lets you choose a task, chat, note, folder or entire Workspace. A set can include several tasks, a note and a calendar.
Where does the assistant get its answers?
Browse sources just as you would in the Files module: from a Workspace to a thread and then to its folders. Click a row to select a source, or click the “›” arrow to go one level deeper. When you select a task or chat, the assistant also gets the files linked to it - both message attachments and files from the Files module. The number of files is shown next to the thread name.
The bar below the list shows how much knowledge fits in the prompt. If you select too much, the assistant will tell you what it left out. It leaves out general areas first, while explicitly selected tasks are retained. A file selected as a source is read in chunks of around 24 000 characters. The assistant can retrieve more, up to 300 000 characters. An attachment read directly from a message has a lower limit - around 50 KB of text.
In the Pro plan, without manually choosing sources, the assistant searches the index built in the background by keywords. With OpenAI or Gemini and semantic search enabled, it also searches by meaning. More recent entries carry more weight. After 45 days, the weight of a chat summary drops by half, and the weight of a recorded decision does so after 180 days. The content of fragments in the index is encrypted, but the vectors are not.
The assistant can only see material you have permission to access. It will not show you a colleague's private note, even if you are an administrator. It checks permissions when reading the source. If you lose access to a source, it will not use it in its next answer.
What do you gain?
A new team member does not have to wait for someone to find the complaints procedure. They can ask the assistant about it, even if it is in a note or file in a folder they can access.
You don't have to go through old conversations to piece together what was agreed in April. Ask about what you recorded instead of looking for someone who remembers the conversation with the customer. You will not see other people's private notes or files just because you work in the same Workspace.
An everyday example
A renovation company is wrapping up the quarter. A customer calls because they remember a different price per square metre. The “Meblex office renovation” task contains forty messages, three versions of the estimate in XLSX and a signed PDF. The person in charge selects the task as a source. Its attachments are included, so they can ask the assistant what rate was agreed.
The assistant can read spreadsheets and messages to help check the rate. If there is too much material, it may ask you to select a specific file. It can read a longer file in successive chunks rather than stopping at the first one. The conversation is saved as a separate thread in the list. The next time a complaint comes in, you can duplicate it along with its set of sources.
Frequently asked questions
Are my documents sent to an external AI provider?
Yes. The content of the selected source is sent in a prompt to the provider chosen by the administrator: OpenAI, Gemini, Claude or Grok. The content of the sections stored in the knowledge base index is encrypted, but the vectors are not.
Can the assistant read my mail?
Only if the administrator enables mail as an AI knowledge source in Settings. It is disabled by default. Once enabled, you can select only your own mailboxes; the administrator cannot select someone else's mail either. Mail is not added to the knowledge base that is indexed in the background.
What types of files can the assistant read?
The assistant can read TXT, CSV, MD, JSON, XML, HTML, DOCX, XLSX, XLS and PDFs with a text layer. It can analyse images visually if the selected model supports them. It cannot read older DOC files or scanned PDFs, because the programme does not include OCR.
Does indexing notes and files use extra AI requests?
No. Notes and files are added to the knowledge base without calling the model. The programme calculates keywords itself. It only summarises tasks and chats.