How rate limits work
Per-API-key bucket. Refills at sustained rate. Burst tolerance ~10x rate for 10 seconds. 429 response when exhausted with Retry-After header.
Receipt issuance vs reads
Both count against rate limit. Writes are slightly more expensive (3 tokens) than reads (1 token).
Anomaly detector calls
Internal; don't count against your rate limit.
Enterprise burst
For Enterprise customers with predictable burst patterns (CI deployment windows, end-of-month evidence pack generation), we configure higher burst tolerance.
Self-hosted rate limits
You operate; no rate limit from our side. Your DB connection pool is the actual ceiling.
Monitoring usage
/admin/api-usage shows your current rate consumption. Set up your own alerts via Datadog / Prometheus on the metrics.