XiaoZhiClaw - XiaoZhi AI Device Integration

🔒 Security

• ✅ No external API keys stored in code
• ✅ All credentials via environment variables
• ✅ No shell command execution
• ✅ WebSocket connections only (no inbound HTTP)
• ✅ Open source and auditable
• ⚠️ Requires Volcengine Doubao API credentials

Overview

XiaoZhiClaw is an OpenClaw channel that connects XiaoZhi AI ESP32 hardware devices to OpenClaw agents, enabling real-time voice interaction.

Permissions

Required Permissions

• ✅ Network Access: WebSocket server (port 8080 by default)
• ✅ Audio Processing: Opus encoding/decoding
• ✅ STT/TTS API: Volcengine Doubao (HTTPS)
• ❌ No Admin/Root Privileges Required
• ❌ No System Command Execution

Data Flow

XiaoZhi Device → WebSocket → STT (Doubao) → OpenClaw Agent
     ↓                                          ↓
  Microphone                              AI Response
     ↓                                          ↓
  Speaker ← WebSocket ← TTS (Doubao) ← OpenClaw Agent

Use Cases

1. Voice Conversation

Talk to your AI assistant through XiaoZhi hardware
Ask questions and get voice responses
Real-time voice interaction

2. Hardware Control

Control volume, brightness via MCP commands
Hardware status monitoring
Device management

3. Voice Commands

Voice-activated AI assistant
Hands-free operation
Physical AI companion

Usage Examples

Start the Service

# The WebSocket server starts automatically when OpenClaw starts
# Default port: 8080

Configure XiaoZhi Device

Configure your XiaoZhi firmware to connect to:

ws://YOUR_COMPUTER_IP:8080

Voice Interaction Flow

1. User speaks → XiaoZhi microphone captures audio
2. Audio streaming → Opus frames sent via WebSocket
3. STT processing → Volcengine Doubao transcribes to text
4. AI processing → OpenClaw agent processes and responds
5. TTS processing → Volcengine Doubao converts to speech
6. Audio playback → XiaoZhi speaker plays response

Environment Variables

# Required: Volcengine Doubao API Credentials
# Get from: https://console.volcengine.com/
DOUBAO_APP_ID=your_app_id_here
DOUBAO_ACCESS_TOKEN=your_access_token_here

# Optional: WebSocket Server Configuration
XIAOZHI_PORT=8080

# Optional: Audio Configuration
AUDIO_SAMPLE_RATE=16000
AUDIO_FRAME_DURATION=60

Protocol

WebSocket Message Types

Handshake:

{
  "type": "hello",
  "transport": "websocket",
  "audio_params": {
    "format": "opus",
    "sample_rate": 16000,
    "frame_duration": 60
  }
}

Listen Events:

{
  "type": "listen",
  "state": "start"
}

{
  "type": "listen",
  "state": "stop",
  "text": "transcribed text"
}

TTS Events:

{
  "type": "tts",
  "state": "start",
  "text": "response text"
}

{
  "type": "tts",
  "state": "stop"
}

Architecture

XiaoZhi ESP32 ←→ WebSocket Server ←→ OpenClaw Channel ←→ AI Agent
     ↓                ↓                    ↓              ↓
  Microphone    Port 8080          xiaozhiclaw      PocketAI
     ↓                ↓                    ↓              ↓
  Speaker      Opus Audio         Message Router   Response
                     ↓
              Doubao STT/TTS

Notes

1. Network: Ensure port 8080 is open on your firewall
2. Latency: Use wired connection or high-speed Wi-Fi for best results
3. API Credentials: Volcengine Doubao API credentials required for STT/TTS
4. Audio Format: Opus encoding, 16kHz sample rate, 60ms frame duration

Troubleshooting

Connection Refused

• Check if port 8080 is open
• Verify XiaoZhi device network settings
• Check firewall settings

Audio Lag

• Check network latency
• Use wired connection if possible
• Ensure good Wi-Fi signal strength

STT/TTS Not Working

• Verify Volcengine API credentials
• Check API quota and billing
• Verify network connectivity to Volcengine API

Device Not Connecting

• Verify WebSocket URL format: ws://IP:PORT
• Check XiaoZhi firmware configuration
• Ensure OpenClaw gateway is running

Resources

• XiaoZhi AI ESP32 Project
• Volcengine Doubao API
• OpenClaw Documentation
• Opus Audio Codec

Changelog

v1.0.0 (2026-03-12)

• ✅ Initial release
• ✅ WebSocket server implementation
• ✅ Volcengine Doubao STT/TTS integration
• ✅ Opus audio encoding/decoding
• ✅ Real-time voice conversation
• ✅ OpenClaw channel integration

License

MIT License

Author

PocketAI for Leo - OpenClaw Community

Credits

• OpenClaw Team
• XiaoZhi AI ESP32 Project
• Volcengine Doubao
• PocketAI 🧤