Give Your AI Agent a
Real Phone Number.

The open-source voice calling plugin for OpenClaw. Deploy inbound and outbound autonomous AI phone agents in minutes.

View on GitHub Read the Docs

bash

# Download from ClawHub

$ openclaw skills install clawvoice

# Install runtime dependencies

$ cd ~/.openclaw/workspace/skills/clawvoice && npm install

# Configure & go

$ openclaw clawvoice setup

$ openclaw clawvoice call +15551234567

✓ Initializing voice engine...

✓ Connecting via Twilio SIP...

📞 Ringing +1 (555) 123-4567... Call connected.

Production-Ready Voice Infrastructure

Everything you need to give your OpenClaw agent a voice, without the vendor lock-in.

Real Telephony API

Bring your own Twilio or Telnyx API keys. Connect your agent directly to real-world PSTN networks. No intermediary markups, full control over your SIP trunks and routing.

Ultra-Low Latency Voice

Choose your engine. Use Deepgram Voice Agent for blazing-fast ~90ms time-to-first-byte with full human barge-in support, or ElevenLabs Conversational AI for premium, emotionally-aware voices.

Inbound & Outbound

Fully autonomous capabilities. Program your OpenClaw agent to proactively dial out for scheduling/sales, or bind it to a webhook to act as a 24/7 intelligent inbound customer support rep.

SMS & Batch Campaigns

Send and receive SMS alongside voice calls. Run batch calling campaigns from a CSV — with answering machine detection, call recording, and exportable campaign reports.

Post-Call Notifications

Get instant transcripts and call summaries pushed to Telegram, Discord, or Slack. Configure daily call limits and voice profiles per agent.

Voice Memory Sandbox

All call transcripts are isolated in a voice-memory/ directory. Promote data to core memory only after review — blocking audio prompt injection attacks.

Zero-Trust Architecture

Voice Memory Sandbox.
Protect against Prompt Injection.

Voice interfaces open entirely new attack vectors. Malicious callers can attempt audio prompt injections to manipulate your agent's state or access sensitive data.

ClawVoice solves this with the Voice Memory Sandbox. All phone call logs, transcripts, and audio data are strictly isolated in a voice-memory/ namespace. Data requires explicit manual review or secondary LLM sanitization before being promoted to the agent's core memory.

Strict namespace isolation prevents core state corruption.
Configurable auto-redaction for PII and credit card numbers.
Human-in-the-loop approval workflows via OpenClaw Dashboard.

voice-memory/ isolation

# All call data lands in an isolated directory
# Separate from your agent's core memory
voice-memory/
  ├── inbound/
  │   ├── call-2026-03-25-14h32.json
  │   └── call-2026-03-25-16h01.json
  └── outbound/
      └── campaign-march-batch.csv

# Review transcripts, then promote to core memory
$ openclaw clawvoice promote call-2026-03-25-14h32
✓ Transcript sanitized and merged to agent memory.

Get Started in 60 Seconds

Download from ClawHub, install dependencies, add the plugin config, and you're live.

Download from ClawHub

$ openclaw skills install clawvoice

Install runtime dependencies

$ cd ~/.openclaw/workspace/skills/clawvoice && npm install

Bundled deps are not included in the ClawHub package — this step is required.

Add plugin config

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.

Run the setup wizard