Using the API
Rate limits
Per-key request limits and handling 429s.
Each key has a token bucket: it can burst up to its capacity, then refills at a steady rate. Limits are per key, not per account.
| Key | Burst | Sustained |
|---|---|---|
sk_live_ | 600 requests | 10 / second |
sk_test_ | 60 requests | 1 / second |
Over the limit, the API returns 429 rate_limited with retry_after (seconds) in the body. Nothing is charged. Wait that long and retry.
The platforms themselves are rate limited too. We absorb that with caching, retries and failover, and if a platform still refuses you get
platform_unavailable, refunded.