Limits and errors
Errors use the OpenAI format, {"error": {"message", "type", "code"}}, with the status codes below.
Parallel requests
Each account runs up to 8 requests at once. A ninth waits up to 5 seconds for one to finish, then returns 429 concurrent_request_limit. Open a support ticket or talk to us to raise your limit.
Status codes
| Status | Code | Meaning |
|---|---|---|
| 400 | model_required | The request has no model, or its body is not valid JSON. |
| 401 | invalid_api_key | The API key is missing, wrong, or revoked. |
| 402 | insufficient_credits | Your balance is too low. Top up and retry. |
| 404 | model_not_found | The model id is not one of the active models. |
| 413 | request_too_large | The request body is over 64 MiB. |
| 429 | concurrent_request_limit | Your account already has 8 requests running. |
| 429 | rate_limited, rate_limit_exceeded | Too many requests. Wait for the Retry-After header, then retry. |
| 503 | under_capacity | The model is temporarily unavailable. Retry shortly. |
Retrying
Retry 429 after the Retry-After header and 503 after a short wait, with backoff. Don't retry 400, 401, 402, or 404; fix the request or the account first.
The OpenAI and Anthropic SDKs already retry 429 and 5xx with backoff. Raise their retries for agents that run many requests at once.
client = OpenAI(
base_url="https://api.silkcompute.com",
api_key=os.environ["SILK_API_KEY"],
max_retries=5,
)