What's the problem with pasting into ChatGPT?
Teams increasingly ask ChatGPT or Claude for help: summarising agreements, drafting a reply to a client, planning the week ahead. For the model to have something to work with, someone has to copy a task's content, a few emails and part of a chat into the conversation. That takes time, it's easy to miss an important message, and no one knows who pasted what to whom.
In the Pro variant of Task Manager, anyone can link their account to ChatGPT or Claude. The model searches for and reads the content it needs by itself, but only what that person can see in the app.
What does linking an account to ChatGPT or Claude look like?
An administrator switches on AI Connections in Settings once. After that, everyone copies the Task Manager address from their Profile and adds it in ChatGPT or Claude as a custom connector. Your Task Manager's login page opens: you sign in, choose the connection validity (90 days by default), and click Allow.
From that moment on, whenever the model needs data, it searches for it in tasks, chats, notes, files, mail from your inboxes and the AI Knowledge Base, and in the calendar it checks due dates, events and team absences. It reads the full content of the items it finds, but doesn't add, change, send or delete anything.
What do you gain?
You don't paste content into the conversation with AI. You simply ask, and the model fetches the task, email thread or note itself.
The model only sees what you see. Permissions are the same as in Task Manager, so an employee won't learn anything through ChatGPT that they can't see in the app.
You're in control of the connections. Anyone can disconnect their own connection in their Profile, and an administrator can see all connections and a reading log: who, when, which app, and what was read, without the content itself.
You can safely add more inboxes. An inbox set to Read-only won't send anything from Task Manager, and the linked model still knows its correspondence.
Use case: office work and a question about agreements with a client
Ola handles a dozen or so clients in the office and works with ChatGPT every day. A client calls asking about a change in the scope of work. Ola asks ChatGPT: "What did we agree with Kowalski about the scope, and what did they write most recently?". The model searches for "Kowalski" in Task Manager, reads the task with the agreements and the latest email thread, then summarises the key points. Before Ola proposes a meeting to the client, she also asks: "What have I got next week?" - and sees her own schedule and who on the team is on leave at that time.
Ola asks for a draft reply. ChatGPT writes it in the conversation, and Ola pastes the text into the thread and sends it herself, because the model can't send anything from Task Manager. On Friday, the administrator sees in the log when Ola's connection searched for data and which items it read.
Frequently asked questions
Are AI Connections included in the Standard variant?
No, this is a Pro plan feature.
Can the model send an email or change a task?
No. The connection is read-only, so the model won't add, change, send or delete anything.
Could an employee see someone else's matters through ChatGPT?
No. The model sees exactly what that person sees in Task Manager, with the same permissions.
Where does the content that's read end up?
With the AI app provider: OpenAI for ChatGPT and Anthropic for Claude, including outside the European Union. That's why an administrator switches the feature on knowingly.