ClawVoice

Give your AI agent a real phone number. It answers calls, makes calls, sends texts, takes messages, books appointments, and reports back to you — all autonomously.

ClawVoice turns your OpenClaw agent into a phone-capable assistant. Say "Call the dentist and schedule a cleaning for next week" and your agent handles the entire call, then sends you a summary on Telegram with the full transcript.

Website: clawvoice.io

Why ClawVoice?

OpenClaw already has @openclaw/voice-call for PSTN calling. ClawVoice is a community alternative that adds batch operations, SMS, post-call intelligence, and a guided setup experience.

vs @openclaw/voice-call (official plugin)

| Capability | @openclaw/voice-call | ClawVoice | |------------|:----------------------:|:---------:| | Outbound PSTN calls | ✅ | ✅ | | Inbound call handling | ✅ | ✅ | | Multi-turn voice conversations | ✅ | ✅ | | Twilio | ✅ | ✅ | | Telnyx | ✅ | ✅ | | Plivo | ✅ | — | | SMS send/receive | — | ✅ | | Batch calling + campaign reports | — | ✅ | | Post-call notifications (Telegram/Discord/Slack) | — | ✅ | | Voice profiles (owner identity per call) | — | ✅ | | ElevenLabs Conversational AI agent | — | ✅ | | Deepgram Voice Agent | — | ✅ | | Setup wizard + health diagnostics | — | ✅ | | Call state machine (13 states) | ✅ | — | | Tailscale Funnel exposure | ✅ | ✅ | | Latency analysis CLI | ✅ | — |

Choose @openclaw/voice-call if you need Plivo support or prefer the official plugin with OpenClaw core team maintenance.

Choose ClawVoice if you want SMS, batch campaigns with CSV reports, post-call notifications, voice profiles, or a wizard-guided setup.

Note: OpenClaw also has Talk Mode — a built-in local voice chat for speaking with your agent through your device's microphone. Neither ClawVoice nor @openclaw/voice-call replaces Talk Mode; they add PSTN telephony (real phone calls to real phone numbers).

Features

| Feature | Description | |---------|-------------| | Outbound calls | Agent places calls with full context about purpose, owner identity, and callback number | | Inbound calls | Agent answers your phone number, takes messages, screens callers | | Batch calling | Sequential calls from a list (up to 20), with consolidated reporting | | Campaign reports | CSV spreadsheets with extracted details, summaries, and full transcripts | | SMS send/receive | Text messaging with auto-reply for inbound texts | | Post-call notifications | Rich summaries to Telegram/Discord/Slack with transcript file attachments | | Voice profile | Owner identity, phone number, and preferences loaded into every call | | Memory isolation | Voice interactions sandboxed from main agent memory | | Safety guardrails | AI disclosure, call duration limits, tool restrictions, answering machine detection | | Dual voice engines | ElevenLabs Conversational AI or Deepgram Voice Agent |

Legal Notice: Automated calling is subject to the TCPA (Telephone Consumer Protection Act) and state telemarketing laws. You are responsible for compliance including obtaining proper consent. Batch calling features are provided as-is — use at your own risk. This software does not provide legal advice.

---

Getting Started

There are two ways to set up ClawVoice:

| Method | Best for | |--------|----------| | Guided setup — Tell your agent "Set up ClawVoice" or run openclaw clawvoice setup | First-time users. The wizard walks through every step with explanations. | | Manual setup — Follow the steps below | Experienced users or automated deployments |

What You'll Need

| Requirement | Where to get it | Cost | |-------------|----------------|------| | OpenClaw installed and running | openclaw.dev | Free (open source) | | Phone number from Twilio or Telnyx | twilio.com or telnyx.com | ~$1.50/mo | | Deepgram API key | deepgram.com | Free tier available | | ElevenLabs API key + Agent ID (optional, for premium voices) | elevenlabs.io | Free tier available | | Tunnel tool (ngrok, Cloudflare Tunnel, or Tailscale Funnel) | See Step 2 | Free options available |

Note: A Deepgram API key is always required — even if you use ElevenLabs for voice. Deepgram handles speech-to-text for call transcription.

Cost per call:

| Voice Stack | Telephony | Voice AI | Total per minute | |-------------|-----------|----------|------------------| | Deepgram (recommended to start) | ~$0.01 | ~$0.01 | ~$0.02/min | | ElevenLabs (premium voices) | ~$0.01 | ~$0.12–0.15 | ~$0.13–0.16/min |

A typical 5-minute call costs $0.10 on Deepgram or $0.65–0.80 on ElevenLabs. Both voice providers offer free tiers to get started.

If you're using ElevenLabs: Create your ElevenLabs Conversational AI agent before running the setup wizard — the wizard asks for your Agent ID. See Step 4 for instructions.

If you already have @openclaw/voice-call installed: ClawVoice replaces the built-in voice plugin. Disable it to prevent conflicts: openclaw plugins disable voice-call. The setup wizard will detect and warn you about this automatically.

1. Install

There are three ways to install ClawVoice. Option A is recommended for most users.

Option A: ClawHub + config (recommended)

Download ClawVoice from ClawHub, install its dependencies, then register it as a plugin via config:

# Download from ClawHub
openclaw skills install clawvoice

# Install runtime dependencies (required — bundled deps are not included in the ClawHub package)
cd ~/.openclaw/workspace/skills/clawvoice && npm install

Then add ClawVoice to your openclaw.json (usually ~/.openclaw/openclaw.json):

{
  // ... existing config ...
  plugins: {
    enabled: true,
    allow: ["clawvoice"],
    load: {
      paths: ["~/.openclaw/workspace/skills/clawvoice"]
    },
    entries: {
      clawvoice: {
        enabled: true
      }
    }
  }
}

Restart your gateway for the plugin to load. Verify with:

openclaw clawvoice --help

Why not openclaw plugins install? The built-in security scanner flags ClawVoice's legitimate use of environment variables (reading API keys) combined with network calls (sending them to Twilio/ElevenLabs/Deepgram) as "possible credential harvesting." This is a false positive — ClawVoice needs to read your API credentials and send them to your configured providers. The plugins.load.paths approach in openclaw.json bypasses the install scanner entirely while still loading the plugin normally.

Option B: Install from source (for contributors or pre-release versions)

git clone https://github.com/ClawVoice/clawvoice.git
cd clawvoice
npm install && npm run build

Then add to openclaw.json as shown in Option A, replacing the path with your clone location:

{
  plugins: {
    enabled: true,
    allow: ["clawvoice"],
    load: {
      paths: ["/path/to/clawvoice"]
    },
    entries: {
      clawvoice: { enabled: true }
    }
  }
}

Option C: npm global + config

npm install -g clawvoice

Then add to openclaw.json using the global npm path. Find it with npm root -g:

{
  plugins: {
    enabled: true,
    allow: ["clawvoice"],
    load: {
      // Use the output of: npm root -g
      paths: ["/usr/lib/node_modules/clawvoice"]
    },
    entries: {
      clawvoice: { enabled: true }
    }
  }
}

2. Start a Public Tunnel

Twilio/Telnyx need to reach your machine from the internet. ClawVoice runs a standalone server on port 3101 that handles both webhooks and media streams.

# Option A: ngrok (quickest)
ngrok http 3101

# Option B: Cloudflare Tunnel (stable URL, free)
cloudflared tunnel --url http://localhost:3101

# Option C: Tailscale Funnel (if you use Tailscale)
openclaw clawvoice expose --mode funnel
# Or manually (path-based):
tailscale funnel --bg --yes --set-path /media-stream http://127.0.0.1:3101/media-stream

The expose command auto-detects your Tailscale DNS name, activates Funnel on the media stream path, and prints the WSS URL to use. Use --mode serve for Tailnet-only (no public internet) or --mode off to disable.

Tunnel URL changes: Free ngrok URLs change every restart. You'll need to update your Twilio webhooks and twilioStreamUrl each time. For a stable URL, use ngrok with a custom domain ($), Cloudflare Tunnel, or Tailscale Funnel.

Cloudflare + Twilio WebSocket: Cloudflare Tunnel has a known issue with Twilio Media Streams. If you get Error 31920, use ngrok instead.

3. Run the Setup Wizard

openclaw clawvoice setup

The wizard walks you through:

• Telephony provider selection (Twilio or Telnyx)
• API credentials
• Tunnel URL configuration
• Voice provider selection (ElevenLabs or Deepgram)
• Voice provider credentials
• ElevenLabs agent configuration (if applicable)

When asked for the Twilio media stream URL, enter your tunnel URL with /media-stream:

wss://YOUR-TUNNEL-URL/media-stream

4. Configure ElevenLabs Agent (if using ElevenLabs)

This step is critical — without it, your voice agent won't know why it's calling or who it represents.

On the ElevenLabs Dashboard:

1. Create a Conversational AI agent (or use an existing one)

2. System Prompt — must include this dynamic variable placeholder:

   {{ _system_prompt_ }}
   

   ClawVoice injects per-call context through this vari