01 / Dimensions
Limit more than request count.
OpenWaya evaluates limits at the API key, organization, model, route, and plan layers. The tightest applicable limit wins, while admin-configured emergency controls can reduce exposure without a deployment.
- Requests and tokens per minute protect shared capacity.
- Concurrent request and stream caps prevent connection exhaustion.
- Daily and monthly spend controls cap financial exposure.
02 / Headers
Make remaining capacity observable.
Successful and rejected responses expose safe rate-limit headers for the active window. A 429 response includes a stable error code, retryability, request ID, and Retry-After when a deterministic recovery time exists.
Response headers
x-ratelimit-limit-requests: 120
x-ratelimit-remaining-requests: 17
x-ratelimit-reset-requests: 2026-09-05T20:30:00Z
retry-after: 203 / Client behavior
Back pressure beats retry storms.
Queue or shed work before exhausting a limit. Bound client retries, use idempotency keys for retryable mutations, reduce concurrency after 429 responses, and request a plan change only after measuring sustained demand.