Concepts
Rate limits
Predictable ceilings with honest headers.
Default limits are 600 requests per minute per key with bursts of 100, returned on every response. These numbers are provisional while commercial tiers are finalised.Proposed design
Response headers
RateLimit-Limit: 600 RateLimit-Remaining: 597 RateLimit-Reset: 42 # seconds Retry-After: 12 # only on 429
Treat 429s as back-pressure: pause the worker for Retry-After seconds and resume. Idempotency keys make the retry safe.