Start Here / Rate Limits
Rate Limits
Per-token and per-account limits, the burst allowance, and what to do on a 429.
Limits are per token, with a fixed ceiling across all of an account's tokens so twenty tokens cannot add up to twenty times the budget. GET /me reports the numbers a token is held to.
| Class | Sustained | Burst |
|---|---|---|
| Reads | 300 per minute | 600 |
| Review writes | 60 per minute | 120 |
| Agent messages | 30 per minute | 60 |
| Whole account | 900 per minute | 1,800 |
Over the limit is 429 with a Retry-After header in seconds. A wall screen polling /review every 10 seconds is well inside.
When the limiter itself is down
Reads still proceed, because a read cannot spend money or send anything and a wall dashboard going blank during a blip is the wrong trade. Writes and messages return 503, never a fake pass, because the ceiling they rely on is the thing that broke.
Agents: this page is also markdown at /docs/rate-limits.md, answers Accept: text/markdown, and is listed in /docs/llms.txt.
Frequently Asked Questions
No. Each class has its own bucket per token, plus the account-wide ceiling. Heavy polling cannot starve approvals.