Usage only by default — never what anyone types, unless a workspace turns content mode on.
IA4AI shows a company which AI chat tools its people use and how much, so it can review its AI spend and stop paying for tools nobody uses.
The extension observes requests to AI chat sites (ChatGPT, Claude, Gemini, Microsoft Copilot, DeepSeek, Kimi) and reports, per person:
The optional gateway proxy for coding agents records, per call: model, provider, token counts, cost, and tool-call names — metadata, never the content.
By default, nothing anyone types, and nothing any model answers. Prompts, replies, file contents and tool arguments are never read, stored or sent. Lengths, counts, timings and code words are the limit.
No prompt, reply or tool argument is captured unless a workspace turns content mode on — and only that workspace's own administrator can, for a window of 1 to 90 days, choosing the scope and the kinds:
To report AI usage to the company that runs IA4AI — totals per tool and per person, seat and spend review. Nothing is sold, and nothing is used for any unrelated purpose.
The company sets the retention period for browser usage events (default 90 days, maximum 400). People can export or erase their own events; the company's data-protection contact handles those requests. See the data-processing agreement for the full terms.
The company operating this IA4AI server is the data controller. Its contact is the person who invited you.