Appendix C. Error codes and handling suggestions
Full-length chapter. Content is checked against the 2026.10 mainline code and the live console; menus and features depend on the current license, account permissions, and delivered environment.
Back to overview · Full table of contents · Download full PDF
The table lists common errors the product actually returns. When troubleshooting, look at the HTTP status code first, then at code and reason in the response body.
| HTTP | code / reason | Meaning | What to do |
|---|---|---|---|
| 401 | invalid_api_key / LLM_AUTH_FAILED | The credential is invalid, revoked, or expired | Confirm the key status and validity in 「API key management」; reissue it and update the caller configuration |
| 403 | Insufficient permission or scope | The caller is not in the available scope of a group that contains this model, or a traffic rule rejected the call | Check available scope; 「Routing & changes › Hit records」 in the request detail names the rule that matched |
| 403 | CONTENT_POLICY_BLOCKED | Content was blocked by a content-check rule; message is the configured reply | Check the matched items in the rule detail traffic view; if it is a false positive, adjust the rule or switch to 「Record only」 first and observe (section 14.12) |
| 403 | CONTENT_SAFETY_UNAVAILABLE | The judge model is unavailable, and the rule's 「When the judge model is unavailable」 is 「Reject」 | Check the judge model status in 「Global settings › Default models」 or on the rule canvas, and change the model if needed (section 14.3, section 14.9) |
| 403 | REQUEST_ORIGIN_REJECTED | The origin of a control-plane write request failed CSRF checks | Operate from the console pages as usual; cross-site or forged-origin requests are rejected |
| 404 | model_not_found | The requested model is not in the services available to this key | Check the model name and the key's service; confirm the model has been added to the service and published |
| 404 | unsupported_model_endpoint | The requested path is not a standard model endpoint | Use a standard endpoint; for pass-through as-is, use 「API services」 |
| 413 | Request body too large | The request body exceeds the limit set on the rule | Shrink the body, or have an administrator raise 「Body limit」 |
| 429 | rate_limit_exceeded / TRAFFIC_RATE_LIMITED | A request-rate or concurrency limit was hit | Wait for retry_delay_seconds in the response, then retry; raise the limit if a higher quota is needed |
| 429 | TRAFFIC_CONCURRENCY_LIMITED | Concurrency exceeded the cap | Lower concurrency, or raise the concurrency cap |
| 429 | TRAFFIC_QUEUE_TIMEOUT | Queue wait timed out | Lower concurrency; or change the overflow action from 「Queue and wait」 to 「Reject immediately」 so the call fails fast |
| 429 | insufficient_quota / POOL_EXHAUSTED | The token or money budget is exhausted | Wait for the period to reset, or adjust the quota; a quota of 0 is a hard wall |
| 503 | TRAFFIC_CIRCUIT_OPEN | The circuit breaker is open | Wait for the open duration to end and watch half-open probes; also check whether the upstream keeps failing |
| 503 | model_route_unavailable / MODEL_ROUTE_NOT_CONFIGURED | No usable route target, or routing is not configured | Check that the model group is published and that the model's upstreams or the rule targets are healthy |
| 503 | token_rate_limit_backend_unavailable | The limit-counting backend is unavailable | This is a platform-side issue; contact support |
| 504 | LLM_TIMEOUT | Upstream timed out | Check upstream availability and timeout settings; a header timeout does not replay the request |
| 502 | LLM_STREAM_INTERRUPTED | A streaming response was interrupted mid-way | The client should treat it as an interrupt; check upstream stability |
| 400 | VALIDATION_FAILED / invalid_request | The request structure does not match the API | Check fields and types against the API docs; when the error comes from gateway-side validation, the request is not forwarded upstream |
Note: under the same HTTP status code, a different reason means a different cause and a different fix. For example, among 429s, TRAFFIC_RATE_LIMITED is rate limiting and POOL_EXHAUSTED is quota exhaustion. Do not look at the status code alone.
Previous: Appendix B. Common task index · Next: Appendix D. Capability notes
Appendix B. Common task index
Full-length chapter. Content is checked against the 2026.10 mainline code and the live console; menus and features depend on the current license, account permissions, and delivered environment.
Appendix D. Capability notes
Full-length chapter. Content is checked against the 2026.10 mainline code and the live console; menus and features depend on the current license, account permissions, and delivered environment.