Skip to main content
Current default authenticated limits are measured per endpoint, independently for the business and API key. Authentication also has separate IP and credential limits. A request must pass every applicable limit.
The current implementation does not set the standard Retry-After response header. Read retryAfterSeconds from the JSON body until the public header contract is decided.
Wait at least the indicated number of seconds, add random jitter, then retry. Cap concurrency and backoff centrally so parallel workers do not create a retry storm. Rate-limit storage fails closed; an unavailable limiter may return api_key_rate_limit_unavailable rather than permitting unmetered traffic.