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

# Rate limits

> When you get rate limited, what's returned, and how to retry gracefully

To keep the service stable and prevent abuse, FortAPI rate-limits requests. When triggered it returns **HTTP `429`** (Too Many Requests). Normal usage rarely hits this — rate limiting mainly catches runaway loops, scrapers, and attack traffic.

## Kinds of rate limiting

* **Gateway / account-level limiting**: caps request frequency per source; returns `429` when exceeded. This stops any single source from saturating the whole service.
* **Upstream provider limiting**: upstreams (OpenAI / Anthropic, etc.) have their own rate limits. When hit, their `429` (or Anthropic's `529` overload) is **passed through verbatim** by FortAPI, with `type` like `rate_limit_error` / `overloaded_error`, and per-channel automatic retry where possible.

<Note>
  Rate limits cap **frequency** (requests per unit time) — separate from your **quota/balance**. Insufficient balance returns `403`, see [Errors](/en/errors).
</Note>

## How to avoid / handle it

1. **Retry with exponential backoff**: after a `429`, wait and retry with progressively doubling intervals (1s, 2s, 4s…) plus a little jitter so clients don't all retry at the same instant.
2. **Don't busy-poll**: cap concurrency on batch jobs and add intervals inside loops.
3. **Reuse connections / batch requests**: combine work into a single request where the API supports batching.
4. **Multiple keys are not a bypass**: limits key off the real source, so spinning up extra keys won't multiply your throughput and may trigger abuse controls.

## Need higher throughput?

If your workload genuinely needs high concurrency and you're being throttled, email [support@fortapi.com](mailto:support@fortapi.com) with your use case and we'll review. Your overall request ceiling is governed by your **account / token quota** (see [Billing & quota](/en/billing)); frequency limits are just a protective guardrail.