A second ingress guard limits requests to 120 per minute per client address per API instance by default, including failed authentication and credential management requests. It may apply to multiple clients that share a network/proxy address. Do not assume all 429 responses include company quota headers.
The company quota is stored atomically in PostgreSQL and survives application restarts. Requests that pass authentication consume quota even if their operation later fails validation or permission checks.
Wait for
Retry-After, add a small random delay, and retry with exponential backoff. Cache stable metadata and use updated_since filters instead of repeatedly fetching your full history.