On Monday, Marta, the owner of an 11-person accountancy firm, pauses before entering her API key. She wants to know whether the company providing the model will see clients' invoices, mail and the entire task database. Marta wants to find out what goes to the AI provider with her question, what stays on the company's server and whether the model receives content from outside the conversation window too.
What is sent to the AI provider when you ask a question?
Your question and the context the AI Assistant uses to answer it are sent to the AI provider. This does not mean the entire company database is sent, but it also does not mean that its contents never leave the server.
Task Manager runs on your own server: the PHP files and MySQL database are stored there. This is where you store your application data, although questions sent to the model leave that server. Your browser does not download libraries, fonts or icons from external servers; outgoing connections are initiated by the application server. The server may also connect to a mail server or the licence panel if you enable these connections. When you use the AI Assistant, the server sends your question and the context it needs to the model you selected. This distinction matters more than the phrase “your data stays with you”.
You can choose OpenAI, Google Gemini, Anthropic Claude or xAI Grok. You provide your own API key - an identifier that lets the application use your account with the provider - and pay that provider directly. Task Manager does not add a markup; your saved key is encrypted and is never displayed in full. Check the selected company’s data processing terms before you start sending it company content. Read more about how costs and responsibilities are divided in our article on using your own AI key in a business.
Can the Assistant only see sources you select manually?
No. Manually selected sources guide the conversation towards the topics you specify, but the AI Assistant can also find a document in the knowledge base that you can access without you adding it using the Sources button. A summary of your activity history may also be included in the Assistant’s context if you use it.
A new conversation thread starts as a general one and can access your tasks. The “Sources” button lets you add a calendar, Workspace, a specific task or chat, a note, folder or file. Mail only appears as an option after an administrator gives permission. Each thread has its own set of sources, so you do not have to use the same scope for every conversation.
On Tuesday, Marta creates a thread called “Month-end close” and selects her team’s Workspace. She separately starts a thread called “Nord client”, where she selects one task and a meeting note. The difference is practical: the first question may need a broader view of the work, while the second does not need one just because it concerns a client.
When selecting a task or chat, check the attachments too. The Assistant also uses files added to that thread, both from messages and from the Files module; their number is shown next to the name. The usage bar below the source list shows how much of the selected information fits. If you select too much, the Assistant will tell you what it left out. It retrieves context from the database with every question, so a newly added attachment may be included in the next answer.
Does a private answer mean the content does not leave the company?
No. “Private” here describes who can see the conversation in Task Manager, not where the model processes the question. The question and the context used are still sent to the AI provider.
On Wednesday, Marta opens a task and clicks “AI Assistant” next to the reply field. In this mode, the Assistant can see the thread content and its attachments, while her question and the answer are hidden from the other participants. The answer is marked “ONLY YOU”. Separate conversations with the Assistant are private in the same way - not even an administrator can see them.
When you click “Organise”, you choose one of four thread reading modes: “Original content”, “Polish”, “Summarise” or “Expert”. Messages are processed in batches of 8, starting with the latest; the result is available only during the session, and the originals remain unchanged. The message content is also sent to the AI provider when you use this feature, even if you don't open a conversation with the AI Assistant.
Permissions set a boundary, but do not replace choosing sources
The AI Assistant can only access tasks and chats you have permission to see; an administrator can access all tasks and chats, but not other people’s conversations with the AI Assistant, notes that have not been shared, or someone else’s Mail. Choose sources separately for each thread, because having permission to access a task or chat does not mean its content is needed for every question.
The knowledge base works in a similar way: Notes and Files retain their own permissions. Assigning a note to a Workspace does not, by itself, share it with the team. If you lose access to a source you previously selected, the Assistant will leave it out the next time you ask a question. That is one boundary; the other is your decision about what scope to ask the model about. Our guide to roles and permissions for tasks explains this in more detail.
Which four settings help narrow the flow of data?
Start with AI permissions, Mail access, the attachment summarisation mode and your own activity history. They are not a single privacy switch: each limits a different type of context or model request.
| What to check | How it affects AI use |
|---|---|
| AI permissions for tasks, chats and the Assistant window | The administrator grants these separately. Without the relevant permission, the feature disappears from the user interface. |
| “Allow Mail to be used as an AI knowledge source” | This option is off by default. Once enabled, a user can select only their own mailbox or thread; turning it off again also disconnects previously selected Mail sources. |
| “AI attachment summaries” | The default “Manual” mode requires you to click “Create summary”. Each file summarised counts as one API request; automatic mode runs in the background and requires cron - a recurring task on the server. |
| Activity history in Profile | You can turn off activity logging, exclude selected areas or use “Clear history”, which deletes the log and the summary prepared for the Assistant. |
Why is Mail switched off by default? Correspondence also contains third-party data, and the contents of a selected mailbox or message may be sent to an external model. Even when this option is enabled, an administrator cannot see someone else’s Mail. Mail is also never added to the knowledge base indexed in the background.
Do not confuse activity history with the history of conversations with the Assistant. An administrator can separately enable or disable saving conversations to the database; saving is enabled by default. This setting controls whether conversations are stored in Task Manager, not whether you send a question to the model. The separate “AI Assistant with internet access” permission allows the content of a specified webpage to be retrieved, but it does not turn the Assistant into a search engine.
The knowledge base can find a document without you selecting it manually
The AI knowledge base helps the Assistant find information in materials you can access, even if you do not add a specific file to the conversation sources. So when asking what the model will see, also consider documents that may be found this way.
Tasks and chats are added to the knowledge base as summaries created by the model. Notes and Files are indexed in their original form, split into fragments; indexing these documents does not require a query to the model. Document indexing does not use up model queries, so it costs nothing and works even when AI is switched off - in that case, searches use keywords. For more on the difference between the index and conversation sources, see the guide to AI knowledge bases and knowledge sources.
By default, the knowledge base is updated once a day; an administrator can choose to update it manually. A deleted document is removed from it at the next update. The knowledge base reads text formats from Files: TXT, CSV, Markdown, DOCX, XLSX and PDFs with a text layer. A scanned PDF without text cannot be read because the programme does not use OCR, or optical character recognition.
Activity History sends a summary, not the raw log
A summary of your Activity History may be included in the AI Assistant's context, rather than the entire log. When you use the Assistant, this summary is part of the information sent to the configured model provider.
The log stores only links and titles - it records things such as a task status change, sending a message or opening a note. It does not record the contents of messages, notes or emails, searches or Automations activity, and Mail is completely excluded from the log; entries older than 90 days are automatically deleted. Only you can see your history, not the administrator.
In your Profile, you can view the 50 most recent events, turn off logging or clear the log along with the Assistant's summary. If you want to exclude just one item, use “Don't save my activity here” in the Workspace, task, chat or note menu. Excluding a Workspace also excludes its tasks, chats and notes, but applies only to you. The app itself generates suggestions from your history, without sending them to the AI provider. Find out more in the Activity History guide.
Where should you start tomorrow, before your team uses AI?
Start by asking about one task in a separate AI Assistant thread; select that task as the source, not the entire Workspace. Then check the answer and the usage bar, review your Mail settings and decide whether you want to use Activity History.
If you're discussing this with your accountant or solicitor, explain the flow without glossing over anything: the app stores data on your server, but questions that include context are sent to the model provider using your API account. Check the provider's terms directly with them; you can also use Task Manager without AI. See how conversations work and how to choose sources in the Demo.
Frequently asked questions
Does my company’s data go to the AI provider?
Yes. When you use the AI Assistant, the provider receives your question and the context used, which may include selected sources, available documents found, or a summary of your activity history. The entire company database is not sent.
Can an administrator read what I ask the AI Assistant?
No. Conversations with the AI Assistant are private, even from administrators, just like your private mail and notes you haven't shared.
Does the AI Assistant have access to the entire company database?
No. Your permissions apply to tasks and chats, while Notes and Files found in the knowledge base retain their own access rules. Administrators have access to all tasks and chats, but not to other people's private AI conversations.
Does the AI Assistant read my mail?
Not by default. Mail can be a conversation source only after an administrator enables the option and you select your own mailbox or thread. The option is disabled by default, and mail is never added to the indexed knowledge base.
What happens if I don't configure AI?
Task Manager continues to work without AI and doesn't send queries to the model provider. Suggestions generated by the application itself based on activity history also don't require an AI provider.
Sources
The Task Manager Team - we write about teamwork, AI in business and keeping tasks organised, using examples from Task Manager, software installed on your own server. Task Manager has been developed by Blue Creation Group (Piotr Łącz) since 2013 and is used by organisations including Neonet, the Military University of Technology and Polpharma - read customer reviews. We check our articles against the user guide and the latest version of the software.
