Core concepts

Rate limits

Per-key limits, the X-RateLimit headers, and handling 429s.

Limits

Requests are limited to 120 per minute per key across the whole surface. Exceeding it returns 429 with the standard error envelope.

Headers

Every response reports your current budget:

Response headers
X-RateLimit-Limit
integer
Requests allowed per window (120).
X-RateLimit-Remaining
integer
Requests left in the current window.
X-RateLimit-Reset
unix seconds
When the window resets.

Handling 429s

A 429 also carries a Retry-After header (seconds). Back off until X-RateLimit-Reset and retry with jitter. For bursty workloads, queue requests client-side, and prefer webhooks over tight polling loops so job status never competes with real traffic for your budget.

In the rare case the rate limiter itself is unreachable, the API fails closed with a 503 service_unavailable; treat it as retryable.