Some routes limit how often one person can call them, because every call costs money. A limit is a sliding window per person (or per token) and route.
Limits today
| Route | Limit |
|---|---|
| Assistant | 20 requests per minute |
| Profile picture upload (in the app) | 10 requests per minute |
The MCP server has no limit of its own yet.
When you are over
The route answers 429 with a Retry-After header holding the seconds to wait, and {"error": "..."} as the body. Wait that long and retry.
How the limit is counted
The counter is kept in the memory of the server instance that handled the request, so separate instances count apart. Read a limit as a brake against a runaway script, not as a quota: the real ceiling can be higher than the table. A shared counter and published quotas for the public API are coming soon.