specification: API Commons Rate Limits
specificationVersion: '0.1'
provider: Green Helix
providerId: greenhelix-net
generated: '2026-09-19'
method: probed
source: Live response headers observed on GET https://api.greenhelix.net/v1/health (2026-09-19) + GET https://api.greenhelix.net/v1/pricing/tiers
sources:
- https://api.greenhelix.net/v1/health (headers)
- https://api.greenhelix.net/v1/pricing/tiers
- https://github.com/mirni/a2a/blob/main/docs/api-reference.md#6-rate-limiting
- https://github.com/mirni/a2a/blob/main/docs/adr/009-auth-rate-limiting.md
description: >-
  Sliding-window per-key limits by tier, with a burst window of (rate_limit_per_hour / 60) + burst_allowance per
  minute, an independent per-tool rate_limit_per_hour where a tool declares one, and a per-IP limiter on public
  endpoints (60 req/min, burst 120 — ADR-009). Exhaustion returns 429 with RFC 9457 code rate_limit_exceeded; a
  wallet budget cap also returns 429 (spend_cap_exceeded). Observed live on an anonymous GET /v1/health: x-ratelimit-limit:
  1000, x-ratelimit-remaining: 995, x-ratelimit-reset: 3103 — the headers are emitted on every response, not only
  /v1/execute and /v1/batch as the api-reference states. No Retry-After header is documented or was observed; the
  docs direct clients to X-RateLimit-Reset (seconds until the window resets) and exponential backoff. A batch of
  N calls counts as N rate events.
headers:
  limit: X-RateLimit-Limit
  remaining: X-RateLimit-Remaining
  reset: X-RateLimit-Reset (seconds until window reset — NOT an epoch)
  requestId: X-Request-ID
  retryAfter: null
observed:
  url: https://api.greenhelix.net/v1/health
  date: '2026-09-19'
  headers:
    x-ratelimit-limit: '1000'
    x-ratelimit-remaining: '995'
    x-ratelimit-reset: '3103'
  note: anonymous request — the limit shown is the unauthenticated/public bucket, not a tier
responseCodes:
  throttled: 429
  spendCap: 429
  codes:
  - rate_limit_exceeded
  - spend_cap_exceeded
limit_count: 6
limits:
- name: Free tier
  scope: per-key
  metric: requests_per_hour
  limit: 100
  timeFrame: hour
  burst: 10 burst allowance (~12/min)
- name: Starter tier
  scope: per-key
  metric: requests_per_hour
  limit: 1000
  timeFrame: hour
  burst: 25 burst allowance (~42/min)
- name: Pro tier
  scope: per-key
  metric: requests_per_hour
  limit: 10000
  timeFrame: hour
  burst: 100 burst allowance (~267/min)
- name: Enterprise tier
  scope: per-key
  metric: requests_per_hour
  limit: 100000
  timeFrame: hour
  burst: 1000 burst allowance (~2,667/min)
- name: Public endpoints per IP
  scope: per-ip
  metric: requests_per_minute
  limit: 60
  timeFrame: minute
  burst: '120'
  applies_to: /v1/pricing, /v1/health, /.well-known/agent-card.json (ADR-009)
- name: Per-tool limit
  scope: per-endpoint
  metric: requests_per_hour
  limit: tool-specific (rate_limit_per_hour in the catalog where declared)
  timeFrame: hour
policies:
- name: Retry
  description: Exponential backoff with jitter on 429/500/503; prefer sleeping X-RateLimit-Reset seconds. 400/401/402/403/404/409
    are non-retryable. SDKs retry 429/5xx automatically.
- name: Batch accounting
  description: POST /v1/batch with N calls records N rate events; balance is checked for the total up front.