Docs Reference
Rate limits
Every rate limit an endpoint applies, as the code enforces it.
Limits are fixed windows: the first request opens a window, and requests past the limit inside it are refused until it closes. Counters are shared across server instances when the database is reachable, and per instance when it is not.
Which limit stops you?
- OverPer IP: 300 per min
- OverPer org: 1,000 per min
1,200 requests, one caller, one window
Refused at #301.
Stopped by the per IP limit, 300 per min. That counter starts again a minute after the request that opened it.
Every limit
| Endpoint | Limit |
|---|---|
| Sending runs | |
/api/ingest | 300 per min per IP1,000 per min per org |
/api/ingest/bypass | 300 per min per IP200 per min per org |
/api/otel/v1/traces | 300 per min per IP1,000 per min per org |
| Runs and replay | |
/api/replay | 30 per min per IP |
/api/runs | 120 per min per org and IP |
/api/runs/[run_id]/bisect | 8 per min per org and IP |
/api/runs/[run_id]/reexecute | 10 per min per org and IP |
| Tests and prompts | |
/api/evals | 6 per min per IP |
/api/evals/pairwise | 6 per min per IP |
/api/prompts/playground | 10 per min per IP |
| Checking records | |
/api/audit/verify | 20 per min per IP |
/api/cps230/verify | 20 per min per IP |
/api/trust/verify | 20 per min per IP |
| Agents | |
/api/agents/authorization-status | 120 per min per org |
Sign-in, billing and this site's own forms are rate limited too (26 more limits), but they are not an API you build against: see API.
Monthly run allowances are per plan: see Plans.