generated: '2026-08-31'
method: searched
source: https://reefapi.com/docs (Rate limits section)
docs: https://reefapi.com/docs
name: ReefAPI
slug: reefapi
limit_count: 3
summary: >-
  Published defaults are ~5 req/s per key and ~10 req/s per IP, enforced over a 3-second
  rolling window so short bursts pass and sustained floods are smoothed. Each engine
  additionally carries its own concurrency budget so one heavy endpoint cannot starve
  the others. Exhaustion returns RATE_LIMITED (429) and the docs say it clears within
  seconds. Limits are raised per account on request.
rate_limits:
- scope: per-key
  limit: 5
  unit: requests
  window: 1s
  enforcement_window: 3s rolling
  burst: true
  burst_note: "Enforced over a short 3-second rolling window — brief bursts are fine."
  status_on_exhaustion: 429
  error_code: RATE_LIMITED
  retryable: true
- scope: per-ip
  limit: 10
  unit: requests
  window: 1s
  enforcement_window: 3s rolling
  status_on_exhaustion: 429
  error_code: RATE_LIMITED
  retryable: true
- scope: per-engine
  limit: null
  unit: concurrent requests
  window: null
  note: >-
    "Each API also has its own concurrency budget, so one heavy endpoint can never
    starve the others." The numeric value is not published per engine — recorded as
    null rather than guessed.
response_headers:
  published: false
  observed: []
  probed: '2026-08-31 POST https://api.reefapi.com/currency/v1/convert (unauthenticated)'
  detail: >-
    No X-RateLimit-*, RateLimit-* (RFC 9331 draft) or Retry-After headers were present
    on the live response. Headers checked on the returned set - date, content-type,
    alt-svc, server, via, cf-cache-status, strict-transport-security, report-to, nel,
    cf-ray. The runtime signal an agent needs is therefore absent - the only rate-limit
    feedback is the body's error.code RATE_LIMITED plus error.retryable true.
  gap: >-
    This is the single clearest agent-readiness gap on an otherwise strong surface.
    Adding RateLimit-Limit / RateLimit-Remaining / RateLimit-Reset (or Retry-After on
    429) would let a client pace itself instead of discovering the wall by hitting it.
quota:
  model: shared credit pool
  free_tier: 1000 credits on signup, no card
  expiry: none (credits never expire; unused monthly credits roll over)
  cost_model: "Most endpoints 1 credit; heavier ones (image rendering, AI, large result sets) cost more, published per endpoint."
  failed_calls: "Failed or blocked calls (ok=false) cost 0 credits."
  exhaustion_code: QUOTA_EXCEEDED (402)
increase_path:
  documented: true
  detail: "Need higher limits for production traffic? Tell us — we raise them per account."
  contact: https://reefapi.com/about