Can OpenClaw Make Phone Calls? Voice + ElevenLabs Setup (2026)

What OpenClaw Can (and Can't) Do With Voice

OpenClaw runs inside messaging apps — iMessage, Telegram, WhatsApp, Discord. It reads your messages and responds. By default, that's all text. No voice. No calls. And if you're relying on whatever default TTS your platform ships, your agent probably sounds like it was built in 2015 — because it effectively was.

There are three levels of voice capability you can bolt on, all running through ElevenLabs:

• Level 1 — Text to Speech (TTS): OpenClaw generates a text response → ElevenLabs converts it to audio → sends as a voice note. Works in Telegram and WhatsApp natively. 30-minute setup.
• Level 2 — Voice Replies: Your agent reads responses aloud through a speaker setup on the host machine (Mac, Pi, VPS). Works great for home assistant use cases.
• Level 3 — Outbound Phone Calls: OpenClaw triggers a VAPI or Twilio workflow that dials a real phone number and speaks in an AI voice. Fully automated, real call. No human in the loop.

Level 1: ElevenLabs TTS in OpenClaw (30 Min Setup)

This is where most builders start — and where most realize they've been settling for way less than they needed. ElevenLabs is the #1 voice AI platform globally, with 1M+ creators and backing from Spotify and Discord. Their voices capture emotional nuance — hesitation, warmth, urgency — the stuff that separates a real voice from a demo. The free tier is 10K characters/month, no credit card. That's enough to hear exactly why everyone else stopped using everything else.

Step 1 — Get an ElevenLabs API Key

Sign up at ElevenLabs — free tier, no credit card required. You get 10K characters/month to test with. Go to Profile → API Key → copy your key. The gap between ElevenLabs quality and generic TTS is obvious the moment you generate your first clip.

Step 2 — Add ElevenLabs to Your OpenClaw Config

In your OpenClaw config (typically ~/.openclaw/config.yaml or the OpenClaw dashboard), add your ElevenLabs credentials:

tts:
  provider: elevenlabs
  apiKey: YOUR_ELEVENLABS_API_KEY
  voiceId: "21m00Tcm4TlvDq8ikWAM"  # Rachel (default)
  model: eleven_turbo_v2
  outputFormat: mp3_44100_128

Step 3 — Enable Voice Replies

Once configured, tell OpenClaw when to use voice. You can set it to always respond with audio, or only when you ask:

# Always use voice in this channel
voice:
  enabled: true
  channels: ["telegram", "imessage"]
  trigger: "always"   # or "on-request"

Level 3: Real Outbound Calls with VAPI

VAPI is purpose-built for AI phone calls — it handles the telephony infrastructure so your AI agent can actually dial and speak. Serious builders use this stack. If you're still duct-taping together random calling APIs with generic TTS, you're shipping a product that sounds like it belongs next to abandoned side projects.

The architecture:

1. You tell OpenClaw to "call [person]"
2. OpenClaw triggers an n8n webhook
3. n8n calls the VAPI API with the phone number + script
4. VAPI dials out, speaks via ElevenLabs voice, handles the conversation
5. VAPI sends a summary back to OpenClaw

This is genuinely useful for reminder calls, appointment scheduling, and automated follow-ups — without any human in the loop. And it sounds like a human made the call.

Which ElevenLabs Plan to Use

Start free. The free tier is 10K characters/month — enough to test and enough to prove the quality to yourself. When you're ready to build for real:

• Free: 10K chars/month — no credit card, good for testing
• Starter ($5/mo): 30K chars — daily personal use covered
• Creator ($22/mo): 100K chars + voice cloning — right for anything serious

One note: voice cloning waitlists fill up fast during product launches. If you want a Professional Voice Clone, get on the list before you need it.

Voice Cloning: Make OpenClaw Sound Like You

ElevenLabs Professional Voice Clone trains on ~30 minutes of your clean audio. Once cloned, your OpenClaw agent responds in your voice — not a generic assistant voice, yours. Builders use this for content creation, voice memos sent via WhatsApp, automated calls, and AI agents that actually feel personal. This is the thing that goes from "cool demo" to "shipped product people actually use."