Every prompt gets an audit trail, and the app didn't change

Compliance wants every AI interaction logged: which container asked, which model, what came back. The application team wants to ship without rewriting the AI client.

Both get it. AI requests on Hoody are plain HTTP to a gateway on your own host, so a hoody-exec script sits in front. It forwards each call to ai.hoody.com, writes the prompt, model, and usage block to an audit store, and returns the response untouched. Turning it on is changing the client's base URL from ai.hoody.com to the exec URL. The container token is the same; the application code is the same.

The log is written by a script your auditor can read, on a server you rent. No vendor dashboard, no SDK hook that changes with the next release.