> ## Documentation Index
> Fetch the complete documentation index at: https://docs.cognizo.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Rate limits

> Limits, the 429 response, and how to back off.

The public API is rate limited per API key. Limits are generous — a normal
integration never hits them — and exist to blunt runaway loops.

## The limits

* **600 requests per minute per key.**
* **600 requests per minute per IP**, counted across every API-key request
  from that address.

Because a key belongs to one organization, per-key is effectively per-org. The
per-IP counter is a backstop: splitting a hot loop across several keys on one
host still trips it.

Both windows are rolling 60-second buckets shared across our web processes —
there's no per-process multiplier to exploit.

## When you exceed it

You get `429 Too Many Requests` with the standard error envelope:

```json theme={null}
{
  "error": {
    "code": "RATE_LIMITED",
    "message": "Too many requests. Please try again shortly."
  }
}
```

## Handling it

* Back off and retry after a short delay — exponential backoff with jitter is
  ideal.
* Prefer bigger pages over more requests — `page_size` up to 100 on most list
  endpoints — instead of fanning out parallel calls.
* Cache what doesn't change per minute. Brand and topic lists rarely move;
  metric windows only shift as new runs land.
* If you legitimately need a higher limit, get in touch — don't work around it
  by rotating keys or IPs.

<Note>
  The [MCP server](/mcp/overview) forwards your key upstream unchanged, so its
  calls count against the same per-key budget as your own integrations.
</Note>

See [Errors](/guides/errors) for the full code list and what's retryable.