Rate limiting across both rings

Public ring: a fixed window per IP

The public ring caps each caller at 60 requests per 60-second window, keyed to the source IP. It is evaluated at the edge, so a throttled request never touches application code. When you exceed the window you get a 429 — wait for the window to roll and continue. For higher sustained throughput you need a partner project.

Partner ring: a token bucket per project

The partner ring meters with a token bucket scoped to your project, not to an individual key or token. Each bucket has a sustained refill rate and a burst capacity; a request costs one token and reads and writes are weighted equally. Because the bucket refills continuously, you can spend your burst in a short spike and settle back to the sustained rate without being locked out for a full minute.

Defaults: build (sandbox) 10 req/s / burst 50; launch 25 req/s / burst 100; scale 100 req/s / burst 400. Sandbox is fixed at build tier regardless of your live tier.

Per-endpoint sub-limits

Some partner endpoints carry their own tighter cap because each call is expensive:

Read the RateLimit-Limit, RateLimit-Remaining and RateLimit-Reset headers on every response and slow down before you hit the wall — see handling rate limits.