Status

What is checked. The gateway probe requests /health over HTTPS and passes when it returns 200. The catalog probe lists /v1/models and passes when the list is non-empty; the count shown is distinct models, not aliases. The upstream probe confirms the inference backend answers at all; a model can still be temporarily unavailable, in which case requests return 503 model_unavailable and the error is deterministic to retry. Probes run from this server, cached for one minute, so timings include one network hop from the site to the API.

If your requests fail while everything here is green, check the error docs first: 401 and 402 are account-side, and 404 model_not_found means the id is not in the catalog.