Browse documentation
Reliability

Rate Limits and Processing Quotas

Design Manifest AI integrations for request limits, submission weights, active-job quotas, and page capacity.

Manifest AI applies layered controls to protect account credits and shared processing capacity: edge traffic controls, per-key weighted request limits, active-job limits, file limits, and plan page capacity.

Rate-limit headers

Authenticated responses report the current weighted request window.

X-RateLimit-Limit
Maximum request units in the current minute.
X-RateLimit-Remaining
Units remaining after this request.
X-RateLimit-Reset
Unix time when the current window resets.
Retry-After
Seconds to wait after HTTP 429.

Request weights

Document submission consumes more rate-limit capacity than a status read.

Cheap reads use one unit. A multipart document submission uses ten units because it performs validation, duplicate checks, encrypted storage, credit reservation, and durable queue creation.

Client behavior

Bound concurrency and use backoff rather than creating a retry storm.

  • Queue submissions in your application.
  • Honor Retry-After exactly.
  • Add randomized jitter.
  • Prefer webhooks over frequent polling.
  • Request Custom limits before a planned volume increase.