Troubleshooting calls
Where to start when a request fails.
Preserve evidence
Record the response status, error body, and X-Request-Id. Do not capture only a fragment of the error message, and do not expose the complete API key.
Troubleshooting order
- Authentication: Does the header belong to the current API family or call entry? Is it a user API key or an upstream key?
- Path: Is the Base URL duplicated with the endpoint path, or is
/v1missing? - Model: Is the public model id published for the current call entry?
- Balance and concurrency: Is the balance greater than 0? Have user, gateway, or upstream concurrency limits been reached?
- Upstream: Use the
request_idto distinguish network errors, rate limits, capacity issues, unavailable models, and timeouts. - Settlement: Check usage, the
estimatedmarker, the usage log, the ledger, and the balance.
See Error codes for common string error codes and Usage and billing for billing rules.
Last updated on