telnyx-voice-sip（telnyx 语音 SIP）

SIP Voice Call Control

Voice interface for OpenClaw using Telnyx Call Control API. Call via SIP and have natural conversations with your AI assistant.

How It Works

Your Phone (SIP) → Telnyx Call Control → Webhooks → This Server → Commands back
                         ↓                              ↓
                    STT/TTS/Media              LLM + Tool Execution

This is not WebRTC — it's server-side SIP call handling. Telnyx handles all audio; we just receive webhooks and send commands.

Setup

1. Install Dependencies

cd skills/sip-voice-call-control
npm install

2. Configure

cp .env.example .env

Edit .env:

TELNYX_API_KEY=KEY_your_api_key_here

3. Start Server

npm run start

Console shows:

🎙️  Assistant Voice Server ready on port 3000
📞 Dial: sip:openclaw@yourconnection.sip.telnyx.com

Agent Setup Flow

When setting up for a user, agents should:

1. Collect Configuration

Ask for:

1. Telnyx API Key (required)

- Get from: https://portal.telnyx.com/#/app/auth/v2 - Needs Voice API + Inference permissions

1. TTS Voice (optional)

- Telnyx.NaturalHD.astra — Female, warm and clear (default, recommended) - Telnyx.NaturalHD.atlas — Male, deep and authoritative - Polly.Amy-Neural — British female - Polly.Brian-Neural — British male - Polly.Joanna-Neural — American female - Polly.Matthew-Neural — American male - Azure.en-US-JennyNeural — American female - Azure.en-GB-RyanNeural — British male - See .env.example for full list

1. Voice Model (optional)

- Qwen/Qwen3-235B-A22B — Best for function calling (default) - meta-llama/Meta-Llama-3.1-8B-Instruct — Fastest - meta-llama/Llama-3.3-70B-Instruct — Balanced

Personalization (assistant name, user name, timezone) is pulled automatically from workspace files (IDENTITY.md, USER.md).

2. Write .env File

cat > .env << 'EOF'
TELNYX_API_KEY=<user_api_key>
VOICE_MODEL=Qwen/Qwen3-235B-A22B
TTS_VOICE=Telnyx.NaturalHD.astra
EOF

3. Start in Background (Persistent)

The server must run persistently to receive calls. Use nohup to keep it alive:

cd /path/to/sip-voice-call-control
nohup npm run start > sip-voice-call-control.log 2>&1 &

Or from an agent:

// Use nohup to keep process alive after session ends
exec({ 
  command: "cd /path/to/sip-voice-call-control && nohup npm run start > sip-voice-call-control.log 2>&1 &",
  background: true 
})

Important: Without nohup, the process will die when the parent session ends. Always use nohup or a process manager for production.

To check if running:

ps aux | grep "tsx.*dev" | grep -v grep

To stop:

pkill -f "tsx.*dev.ts"

To view logs:

tail -f /path/to/sip-voice-call-control/sip-voice-call-control.log

4. Extract SIP Address

Poll the process logs and give the user the SIP dial-in:

📞 Dial: sip:openclaw@<connection>.sip.telnyx.com

Environment Variables

Available Tools

Features

• Low-latency — 500ms-1.5s response time with enable_thinking: false
• Barge-in — Interrupt the assistant anytime by speaking
• Function calling — Native tool support with Qwen
• Auto-setup — Cloudflare tunnel and Call Control app created automatically
• Personalization — Reads IDENTITY.md and USER.md for context

Troubleshooting

No response after speaking:

• Check Telnyx API key has Voice API + Inference permissions
• Verify webhook URL is reachable (tunnel must be active)

Slow responses (>3s):

• Ensure using function-calling branch (not main)
• Check model availability on your Telnyx account

Tool not executing:

• Ensure openclaw CLI is in PATH
• Check WORKSPACE_DIR is set correctly

Port already in use:

• Kill existing server: pkill -f "tsx.*dev.ts"
• Or change PORT in .env

Resources

• Telnyx Call Control: https://developers.telnyx.com/docs/voice/call-control
• Telnyx Inference: https://developers.telnyx.com/docs/inference
• See ARCHITECTURE.md for technical details