# Emotion

> Let the agent laugh, whisper or sound excited where it fits, with ElevenLabs v3.

Source: https://voxio.in/docs/voice-emotion · voxioagent 2.0.0 · Index: https://voxio.in/docs/llms.txt

1. On **Voice**, set speaking to **ElevenLabs** with the `eleven_v3` model.
2. On the AI decision step that writes the reply, turn on **Add emotion** (`emotion: true`) and keep **Emotion style** on ElevenLabs v3 (`emotion_tts: "eleven_v3"`).
3. Speak its `speak` with a Say step as usual. The cues are performed, not read.

```json
"think": {
  "type": "llm",
  "parameters": { "system_prompt": "…", "emotion": true, "emotion_tts": "eleven_v3", "llm_return_type": { "speak": { "type": "str", "description": "What to say." } } },
  "next": "reply"
}
```

> **Warning**
>
> Other voices read cues like `[laughs]` out loud. Only turn emotion on with ElevenLabs v3. AI reply steps don't add cues.

To show a reply without its cues (for example in a transcript), send it through a Say step with `remove: true` and `remove_tts: true`.

## Next steps

- [ElevenLabs](https://voxio.in/docs/voice-speaking-elevenlabs.md): all ElevenLabs settings.