Common problems and troubleshooting

Diagnose using the response

401: check that the key is an active XHub virtual key. 403: inspect model scope, team membership and role. 400: check JSON, required fields, public model name and endpoint protocol; preserve the response body.

Steps

429: inspect RPM, TPM, quota and upstream rate limits. 5xx: use the call ID to locate deployment, upstream errors and timeouts. If curl works but browser requests fail, check gateway URL, HTTPS mixed content and CORS origins.

Checks and limits

GET /models is anonymous deployment discovery; GET /v1/models applies the caller's XHub key permissions. For tool failures, use the Playground gateway root without appending /v1 and inspect paths added by the tool. Preserve x-litellm-call-id, and distinguish media creation, polling and download.