generated: '2026-09-03'
method: searched
source: https://cracked.ai/docs/api (Rate limits section)
limit_count: 3
limits:
  - scope: per-key
    limit: 120
    window: 1 minute
    applies_to: runs (POST /v1/run); discover and inspect are not rate limited; async runs count once, at start
    exhaustion_status: 429
    error_code: RATE_LIMITED
  - scope: per-provider shared token bucket
    limit: provider-published upstream limits (CoinGecko 30/min, Nominatim 1/s, Alpha Vantage 5/min, Hunter 15/s, NCBI 3/s)
    window: per provider
    behavior: >-
      bursts are smoothed - each run waits up to 2s for a token instead of failing; an upstream 429
      with Retry-After <= 5s is retried once; recorded on the run as smoothed {waitedMs, retried,
      retryAfterS} and in the X-Cracked-Smoothed response header
  - scope: per-address (agent self-registration)
    limit: 5
    window: 1 day
    applies_to: POST /v1/agents/register
response_headers: [Retry-After, X-Cracked-Smoothed, X-Cracked-Replayed, X-Cracked-Cost-Micros]
notes: >-
  429 on key exhaustion; capacity-locked providers answer 503 with Retry-After (observed example
  in the docs shows Retry-After: 2712). Key-level monthly_limit_usd caps spend rather than rate.