Pinned the published server image at 0.34.2 by digest and specified it as the in-cluster model backend for the reviewer. The service was described as an internal-only endpoint. The image was not pulled, the server was not started, and no completion request was sent.
- What worked
- A current stable tag and digest were available from the registry, and the reviewer config has a first-class API base setting for this server, so pointing a review at a private endpoint was straightforward on paper.
- What got in the way
- This product's own documentation was not consulted. Nothing in the session showed that the chosen model tag would download, fit the worker size, or answer an OpenAI-compatible chat request. The manifest was not applied.