Rate Limits & Traffic Policing
API-key limits are checked before AI provider work. The key record can define requests per minute (RPM) and requests per day (RPD); configured limits are enforced across the API-key guarded endpoints.
Per-key settings: Configure RPM and RPD from the API Keys page. Changing a key's limits applies to subsequent requests. Empty limits mean no per-key cap for that window.
Standard Operational Limit Bounds
| Metric Target Class | Default Sandbox Threshold | Tracking Logic Interval |
|---|---|---|
| Requests per Minute (RPM) | 60 reqs/min | Counter resets 60 seconds after the bucket reset time. |
| Requests per Day (RPD) | 1,000 reqs/day | Counter resets 24 hours after the bucket reset time; this is not a midnight UTC reset. |
| Monthly Token Volume Limit | Not enforced | A monthly token value can be stored on a key, but it is not currently checked by request handling. |
Limit Exceeded
When a configured RPM or RPD limit is reached, the gateway responds with HTTP 429. The response does not currently include rate-limit headers.
{
"statusCode": 429,
"message": "API key requests-per-minute limit exceeded",
"error": "Too Many Requests"
}
Client guidance: Wait for the documented window to reset before retrying. Use bounded exponential backoff for transient 429 responses.