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

# Rate limits

> Per-key concurrency and the server-wide concurrency budget.

Kova uses **concurrency limits**, not per-minute or per-hour throttles.

## Per-key limit

| Tier                  | Per-key max in-flight | Per-minute / per-hour | Max chars per request |
| --------------------- | --------------------- | --------------------- | --------------------- |
| Standard (`kova_sk_`) | 10                    | no throttle           | no limit              |

If you exceed 10 in-flight requests on a single key, the 11th immediately receives a 429 — it does not queue server-side.

## Server-wide semaphore

The server has a single global semaphore that bounds total concurrent inference (default `MAX_CONCURRENT_TTS=15`). Under sustained heavy load — many keys saturating their per-key limit at once — requests may queue briefly even if your key is under its individual limit.

In practice this is rare; the per-key limit is the binding constraint for almost all integrations.

## Recommended client-side patterns

<CardGroup cols={2}>
  <Card title="Queue locally" icon="layer-group">
    Maintain a bounded semaphore in your application matching the per-key limit. Drop or backpressure once full.
  </Card>

  <Card title="Retry with backoff" icon="rotate">
    On 429, retry with exponential backoff + jitter. Start at 200ms, cap at 5s.
  </Card>

  <Card title="Split across keys" icon="key">
    For higher throughput, create multiple keys (one per service / environment) — each gets its own per-key budget.
  </Card>

  <Card title="Reuse WebSocket" icon="plug">
    For real-time apps, one WebSocket connection counts as one in-flight slot regardless of how many utterances you send through it.
  </Card>
</CardGroup>

## Asking for higher limits

For sustained high-volume traffic, [contact us](mailto:support@kova.ai) to discuss provisioned capacity.