Code review by a model on the same rack as the code

Run an open-weights code model as a daemon in a container on your bare metal server. Point hoody-agent's provider at it over localhost. The agent reads the diff, runs the tests, writes the review. Everything it reads and writes stays on the box. Hoody's API sees a container; it doesn't see prompts, files, or the review.

The public repos take the other path: the Hoody AI gateway, which runs on your host and logs no prompt content, though the upstream provider receives what it's sent. Two providers, two repos, one config line apart.

The reviewer reads the whole repo and the repo never leaves the building.