Skip to main content
Every API key has its own limits, set when the key is created (console, API keys). Every authenticated response reports them in the IETF draft headers:
When a limit is reached the API answers 429 RATE_LIMITED with Retry-After. Organizations have no limit of their own on runs in flight or waiting. If the whole service has too many runs waiting, new runs get 503 SERVICE_UNAVAILABLE with Retry-After, before anything is held; retry with the same Idempotency-Key. See Errors. Spend is limited separately: give each agent its own key with a spend cap (Pricing and billing).