Rate limits
Design clients to handle API quotas and 429 responses.
The currently published limit is 10 requests per second with a burst of 2 and 100,000 requests per day, per API key. A gateway-limited request can return 429 with {"message":"Too Many Requests"} rather than the standard error envelope.
Rate-limit headers and Retry-After are not currently returned. React to 429 with exponential backoff and random jitter. Bound the number of attempts, and use an idempotency key when retrying a write.
Official SDKs retry rate limits and selected transient errors by default. Avoid stacking an unbounded application retry loop on top of SDK retries.
These quotas are operational policy and can change. Treat a 429 as authoritative even if local request counting suggests capacity remains.