# Idempotency

> Retry any create call safely: a repeat never renders or charges twice.

Source: https://developers.mnml.ai/docs/idempotency

> [!TIP]
> **In short:** add an `Idempotency-Key` header that names the work, such as the file name. If the network drops and you send it again, you get the first answer back instead of a second render.

Send an `Idempotency-Key` header with any create call: 1 to 255 visible characters that name the work. If the same key arrives again within 24 hours with the same request, you get the first answer back, byte for byte, with `Idempotent-Replayed: true`. A first answer that was an error replays too.

**Header**

```
Idempotency-Key: facade-042
```

| Case | Answer |
| --- | --- |
| The same request again | The stored answer, with `Idempotent-Replayed: true`. |
| The same key with a different request | `409 CONFLICT`, `reason: key_reused`. Use a new key for new work. |
| The same key while the first is still running | `409 CONFLICT`, `reason: in_progress`. Wait a moment and retry. |
| A retry after a `5xx` | It runs again under the first attempt’s identity, so anything that already started is answered, not started twice. |

- Keys are scoped to your account, so two accounts can use the same key without colliding.
- Use something that names the work, such as the source file or a row id, so a retried batch lines up with the first attempt.
- With no header, the same request from the same key within 10 minutes is still caught and not charged twice. To render the same brief again on purpose, send `count` or a key of your own.
- With `count`, each variation gets its own key derived from yours, so a retry replays them all.