Endpoint
Rate limits
Per-minute, daily, and monthly allowances are shared by every key on the same account and plan. Creating or rotating a key does not multiply capacity.
Response headers
| Field | Required | Notes |
|---|---|---|
| x-rate-limit-limit | per minute | Requests available in the current 60-second window |
| x-rate-limit-remaining | per minute | Tokens left after this reservation |
| x-rate-limit-reset | per minute | Seconds until the current window resets |
| x-quota-daily-{limit,remaining,reset} | daily | Shared daily quota; reset is an ISO 8601 timestamp |
| x-quota-monthly-{limit,remaining,reset} | monthly | Shared monthly quota; reset is an ISO 8601 timestamp |
| Retry-After | 429 | Seconds until the allowance that rejected the request resets |
- The per-minute limiter uses a 60-second window beginning with the first reservation in that window.
- Daily and monthly reservations are serialized at the account level, so concurrent requests cannot all pass a stale quota check.
- MCP and REST batches atomically reserve one token and one quota unit per item before any item runs. A rejected reservation is not partially charged.
- Unlimited daily or monthly quotas use the literal header value unlimited; per-minute limits remain finite.
- A key with no plan is rejected with 403, not treated as unlimited.