Talk Mode & Voice

+15 Mana ✨
OpenClaw Talk Mode(DOCUMENTATION)

Introduction

OpenClaw supports voice interaction through Talk Mode, which enables real-time voice conversations with your agents. Combined with voice wake detection, you can activate your agent with a wake word, have a spoken conversation, and receive spoken responses — all without touching a keyboard.

Key Concepts

  • Talk Mode: An interaction mode where the agent listens for voice input and responds with synthesized speech.
  • Voice Wake Detection: A background listener that activates Talk Mode when a configured wake word is spoken.
  • Voice Overlay: A macOS-specific UI that appears when Talk Mode is active, showing a waveform visualization.
  • Audio/Voice Notes: Support for receiving voice messages from channels like WhatsApp and Telegram, which are transcribed before processing.

Real World Context

A developer is debugging a server issue with both hands on the keyboard. Instead of switching to a chat window, they say the wake word, describe the error verbally, and the agent investigates and responds with a spoken summary. The developer never takes their hands off the keyboard.

Deep Dive

Enable Talk Mode in the agent configuration:

json
{
  "voice": {
    "enabled": true,
    "wakeWord": "hey claw",
    "tts": {
      "provider": "openai",
      "model": "tts-1",
      "voice": "nova"
    },
    "stt": {
      "provider": "openai",
      "model": "whisper-1"
    }
  }
}

This configuration enables voice with the wake word "hey claw". Speech-to-text uses OpenAI's Whisper for transcription, and text-to-speech uses OpenAI's TTS for voice responses.

Activate Talk Mode manually or via wake word:

bash
# Manual activation
openclaw talk

# The agent listens for voice input, transcribes it,
# processes the request, and responds with synthesized speech.

# Wake word detection runs in the background
openclaw voicewake start

The openclaw talk command enters Talk Mode immediately. The openclaw voicewake start command runs a background listener that activates Talk Mode when the wake word is detected.

On macOS, Talk Mode shows a voice overlay:

bash
# The overlay shows:
# - Waveform visualization while listening
# - Transcription text as it's processed
# - Agent response as it streams

Voice notes from messaging platforms are automatically transcribed:

User sends voice note on WhatsApp → 
Gateway transcribes with STT → 
Agent receives text transcript → 
Agent responds (text sent back to WhatsApp)

The transcription happens transparently. The agent receives a text message regardless of whether the user typed or spoke.

Common Pitfalls

  1. High latency with large STT models — Whisper-1 is fast, but larger models introduce noticeable delay. Choose models appropriate for real-time interaction.
  2. Wake word false positives — Short or common wake words trigger accidentally. Use distinctive multi-word phrases.

Best Practices

  1. Use a distinctive wake word — Two or three syllables that don't commonly appear in conversation.
  2. Test in your environment — Background noise levels affect wake word detection accuracy. Test in your actual working environment.

Summary

  • Talk Mode enables real-time voice conversations with agents
  • Voice wake detection activates Talk Mode with a configurable wake word
  • STT (Whisper) transcribes speech, TTS synthesizes responses
  • macOS provides a voice overlay with waveform visualization
  • Voice notes from messaging platforms are automatically transcribed to text
✓ Completed