Rate limits

Every authenticated caller is rate-limited with a token bucket to keep one runaway script from starving everyone else. Limits are enforced per (user, endpoint) — all of your API keys and OAuth tokens share one bucket per endpoint, so minting more keys does not multiply your limit. Anonymous traffic (against free endpoints) shares a per-IP bucket.

Defaults

Limits live on the user's plan. Operators can adjust via the admin plans page; per-endpoint overrides happen at the EndpointCost level (an endpoint that's cheaper or more expensive than the plan default can carry its own ceiling).

Headers on 429

When a request is throttled, we attach:

HTTP/1.1 429 Too Many Requests
Retry-After: 12
X-RateLimit-Limit: 120
X-RateLimit-Remaining: 0
Content-Type: application/json

{
  "error": {
    "code": "rate_limit_exceeded",
    "message": "Per-minute limit exceeded; retry in 12s.",
    "limit_per_min": 120,
    "retry_after_s": 12
  }
}

• Retry-After (seconds) — sleep at least this long before retrying.
• X-RateLimit-Limit — the bucket size for this caller + endpoint.
• X-RateLimit-Remaining — tokens left (always 0 on a 429).
• error.retry_after_s mirrors Retry-After in the JSON body for clients that find headers awkward to read.

Successful (non-429) responses do not carry X-RateLimit-* headers — read your current allowance off the /dashboard/usage/ page instead.

Backing off

Read Retry-After and sleep for at least that many seconds. A simple backoff that handles 429s cleanly:

import time, requests

def call_with_backoff(url, headers, body, max_retries=5):
    for attempt in range(max_retries):
        r = requests.post(url, headers=headers, json=body)
        if r.status_code != 429:
            return r
        delay = int(r.headers.get("Retry-After", "1"))
        time.sleep(delay + 0.1)  # +100ms jitter
    raise RuntimeError("rate-limit retries exhausted")

Behavior during incidents

If the limiter's Redis backend is unreachable, calls fall back to an in-memory bucket per worker process. The limit still applies — it just applies independently per worker, so an N-worker deployment effectively allows N× the configured rate during the incident. You won't be charged extra credits for any over-quota traffic that slips through.

Asking for more

For higher limits than the Pro tier offers — talk to support. Limits are configurable per-customer.