What can an AI assistant see through a connection?

A connected assistant is not a second pair of eyes with its own permissions. It asks as the person who connected it, and every answer is cut to what that person's role already sees in Worktivity.

What it can read

Real employee names, working time, application and website usage, and leave information — all limited by the asking person's role. A manager reaches their own team, an employee reaches only themselves. Cost and pay figures are available to owners only.

What never leaves

Screenshots, timelapse videos and window titles are not on the surface at all. Application usage carries the host name — "github.com" — and never the full address with its path, so aggregated figures travel and row-by-row browsing history does not.

Can it change anything

Only if an owner has turned on the separate write switch, and it is off by default. With it on, an assistant can create and update projects, customers, tasks and task comments, and approve or reject leave requests; each change is recorded in the person's own name. There is no delete tool, and employee records, payroll and billing cannot be written through a connection.

Where the data goes

To the AI provider whose account was connected — Anthropic for Claude, OpenAI for ChatGPT — under that person's own account terms with the provider. Worktivity is not a party to that relationship. This is different from the AI summaries inside the product, where Worktivity sends the request itself and the provider acts as our processor. The privacy policy sets out both.

Every question is written down

Each call is recorded for 180 days: who asked, when, and which tool ran. Owners see this list in Organization settings under AI connections. The wording of the question itself is not stored.

Turning it off

An owner can switch the whole feature off, and every existing connection stops working the same minute. Anyone can disconnect their own assistant from Account settings.