Skip to main content

Limits

  • GET /credits and GET /activity each have one budget per organization: all of your management keys share it, and using one endpoint never spends another’s.
  • GET /key gives every key a budget of its own, so one application checking its key never slows down another.
  • Each window is one calendar minute (UTC): the count starts over at the top of every minute.
  • Every request that passes the key checks counts, including ones refused with 400 or 429. A request refused with 401 or 403 never reaches the limit.
  • These figures change slowly, so one read a minute is plenty for dashboards and alerts. Cache responses rather than polling.

Rate-limit headers

Every response to a request that counts, successful or not, carries the current window:
integer
Requests allowed per minute.
integer
Requests left in the current minute.
integer
When the current minute’s window ends, as a Unix timestamp in seconds.
integer
On a 429 only: seconds until the window ends.

Rate-limited responses

Over the limit, the API answers 429 Too Many Requests with a Retry-After header and this body:

Retry handling

Wait for Retry-After instead of retrying immediately: until the window ends, every request is refused and still counts.

Higher limits

If your workload needs more, email support@pearlresearch.ai with your organization and the endpoints you call.