Errors

Errors use the OpenAI envelope, so the SDKs raise their normal exception classes and your existing error handling keeps working.

{
  "error": {
    "message": "The model `gpt-9` does not exist.",
    "type": "invalid_request_error",
    "param": "model",
    "code": "model_not_found"
  }
}

Status	type	SDK exception
400	invalid_request_error	BadRequestError
401	authentication_error	AuthenticationError
402	insufficient_quota	APIStatusError
403	permission_error	PermissionDeniedError
404	invalid_request_error	NotFoundError
405	invalid_request_error	APIStatusError
410	invalid_request_error	APIStatusError
413	api_error	APIStatusError
422	invalid_request_error	UnprocessableEntityError
429	rate_limit_error	RateLimitError
5xx	api_error	InternalServerError

If you request an unknown endpoint, you'll receive a 404 error with code: "unknown_endpoint". In this case, methods like client.embeddings.create() will raise a NotFoundError instead of a general APIStatusError. If you use the wrong HTTP method on a valid endpoint, the API responds with a 405 status and code: "method_not_allowed".

An unknown model id returns a 404 error with code: "model_not_found".

A retired model id returns a 410 error with code: "model_retired". The message names the model to use instead:

{
  "error": {
    "message": "The model `djelia-asr-v2` was retired on 2026-09-20. Use `sunjata-1` instead. Migration guide: https://docs.djelia.cloud/migration/",
    "type": "invalid_request_error",
    "param": "model",
    "code": "model_retired"
  }
}

A retired native endpoint (/v1/models/*, /v2/models/*, and the same paths under /api) returns a 410 error with code: "endpoint_retired" for every method:

{
  "error": {
    "message": "The native endpoint `POST /v1/models/translate` was retired on 2026-09-20. Use the OpenAI-compatible API at https://api.djelia.cloud/openai/v1 with the model `banjugu-1`. Migration guide: https://docs.djelia.cloud/migration/",
    "type": "invalid_request_error",
    "param": null,
    "code": "endpoint_retired"
  }
}

If you send audio that exceeds the supported size for transcription, the server returns a 413 error and code: "audio_too_large".

An unsupported response_format on jifili-1 returns a 400 error with param: "response_format". The list of supported formats can be found in the Field support documentation.

Sending a text longer than the model accepts triggers a 422 error with code: "input_too_long". The limit is 5000 characters for jifili-1; the message names the limit and the model.

Running out of credit

A 402 means the organisation's wallet lacks the balance for the request. Add credits at console.djelia.cloud.

Errors during a stream

/audio/speech and the streaming responses send their status and headers before the model produces output, so a failure part-way through a stream cannot become a 500. The caller sees a truncated body on a 200 instead.

Nothing is charged for a stream that does not complete.

Validation and Limits

If your request includes invalid fields or values, the server responds with a 400 status before starting inference. For an overview of request limits, see the Field support documentation. When troubleshooting errors, make sure to log the response status, error code, and request ID—but never log your API keys.