Rate limits

Current service protection

API protection is a process-local fixed window of 120 requests per authenticated client per minute, with no queue. Remote IP is the fallback partition when client identity is unavailable. This is not a distributed installation-wide quota or a commercial SLA.

Credentials belonging to the same client share its client partition. Issuing another credential during rotation does not create a separate client quota. Do not design throughput around resetting identity or relying on a new process.

Honor Retry-After

HTTP/1.1 429 Too Many Requests
Content-Type: application/problem+json
Retry-After: 60

Rejection returns rate_limited and conservative whole-second Retry-After. Wait at least the indicated interval. Add appropriate jitter/backoff across multiple workers so they do not all retry together. The example delay is illustrative; read the actual response header.

• Do not poll idle or waiting queues unnecessarily.
• Cache useful task context locally while respecting current write preconditions.
• Retry mutations using their logical key and reconciliation rules.
• Do not infer provider capacity or a guaranteed request rate from this protective window.

See safe retry patterns, rate-limit errors and work discovery.