Rate limits

Two independent limits apply to the data plane. They answer different questions, so hitting one tells you nothing about the other.

The quota is the durable, billable limit — see Usage & quotas. The per-second rate is a burst control that protects shared capacity.

Per-plan rate

These are the defaults. The current values for every plan are on the website’s pricing page and in GET /v1/pricing.

Headers

Every billable data-plane response carries:

• X-RateLimit-Limit — your plan’s per-second allowance.
• X-RateLimit-Remaining — allowance left in the current second.

A throttled response additionally carries Retry-After, in whole seconds.

Throttled response — 429

{
  "code": "rate_limit_exceeded",
  "message": "Short-term request rate exceeded for your plan.",
  "request_id": "...",
  "limit": 2,
  "retry_after_seconds": 1
}

A throttled request is not billed — it never reaches the rate store, so no request unit is consumed.

Handling throttling

Wait for Retry-After and retry. Because the window is one second, a single short sleep is almost always enough; exponential backoff with jitter is still the right shape if you are driving many concurrent workers.

Do not pattern-match on message. Branch on code: rate_limit_exceeded means “slow down and retry”, quota_exceeded means “this billing period is finished — retrying will not help”.