API

API Troubleshooting

Handle API errors consistently and collect useful support information.

Error format

{
  "error": {
    "message": "The `model` field is required.",
    "type": "invalid_request_error",
    "code": "invalid_request_error"
  }
}

Status	Typical code	Action
`400`	`invalid_request_error`	Fix the JSON, model, or messages.
`401`	`invalid_api_key`	Supply a valid active API key.
`403`	`account_suspended`	Check account status or the company seat.
`404`	`model_not_found`	Refresh `GET /models` and use a public model ID.
`409`	`idempotency_in_progress`, `idempotency_key_reused`	Do not run the logical request again. Reconcile the original request ID.
`429`	`quota_exhausted` or request throttling	Honor `Retry-After`; inspect the quota reset time.
`502`	`upstream_error`	Retry with backoff and the same idempotency key.
`503`	`no_route_available`, `overloaded`	Retry briefly or select another capable public model.

Log useful fields

Log the HTTP status, error.type, error.code, public model ID, idempotency key, session ID, attempt number, latency, and X-Censi-Request-Id. Redact the bearer token, prompts, tool results, image URLs, and personal data unless your policy explicitly allows them.

For a streamed response, parse each event because an error can arrive inside an HTTP 200 stream.