Skip to main content

Rate Limits

Rate limiting is per regional account: all API keys of the account share one per-minute request bucket and the monthly request and token quotas, with a per-minute request bucket plus monthly request and token quotas. The burst bucket is local to an instance. Do not infer global multi-instance coordination.

Makas-RateLimit-* describes the burst bucket; Makas-Quota-Requests-* and Makas-Quota-Tokens-* describe monthly dimensions. Limits/remaining are counts; reset fields are Unix seconds. Unlimited dimensions omit their families and unavailable snapshots can omit headers.

A 429 RATE_LIMIT_EXCEEDED is a burst throttle and may carry Retry-After in seconds. TOKEN_QUOTA_EXCEEDED and REQUEST_QUOTA_EXCEEDED are monthly hard stops. A spending cap returns 429 with type: urn:makas:error:cost-cap-exceeded and capScope (account or API key) but no errorCode; it holds until the cap is raised or the month resets. Branch on errorCode when present, otherwise on type; avoid rapid retries on exhausted quota or caps. Provider throttling is separate and its mapped Makas-Provider-RateLimit-* representation follows the provider.

First-launch billing is off: new accounts receive the free Founding plan without a card. Paid tier references describe planned billing-on behavior; plan reservations do not enable paid billing.

Founding includes 5,000,000 requests/month, 200 requests/minute, 3,000,000,000 tokens/month, 10 API keys, 5 team members and 10 custom routing rules; request logs retain 30 days and analytics 90 days. It includes all currently live feature gates except semantic caching. Customer-specific quota overrides can change effective limits.

Complete conditional header reference.