Skip to main content
Calls are limited per account, over all its keys and its console session together. The limits are set per deployment; the numbers here are this API’s current ones.
  • calls per minute: 600, over any 60-second window; past the limit, 429 rate_limited;
  • calls waiting for an answer at once: 10; past the limit, 429 too_many_concurrent;
  • research and resolve jobs still running: 6; every resolve counts, and every circle, edge or paths call with on_miss: RESEARCH, until its job ends, or for at most 2 hours; past the limit, 429 research_busy.
Every 429 carries Retry-After, in seconds. Every answer past authentication carries RateLimit-Limit, the calls allowed per minute. Every /v1 call counts toward the first two limits, /v1/job polls and /v1/openapi.json included; a 429 too_many_concurrent still counts toward the calls per minute. When the rate limits cannot be checked, a call is answered 503 unavailable; repeat the same request. How to stay within them:
  • on a 429, wait Retry-After seconds, plus a small random delay, then repeat the same request: too_many_concurrent sends 1, rate_limited the time until the window has room, research_busy 30;
  • keep at most 10 calls in flight, polls included;
  • polls count toward the first two limits: a job polled as Jobs advises takes about 13 calls in its first minute and 6 a minute after;
  • 429 research_busy means one of the account’s research or resolve jobs must end first; jobs take minutes, so poll the running ones instead of repeating the call in a loop;
  • a 429 reserves and charges nothing.