I’ve been running OpenClaw as my personal AI assistant for a few months now — mostly typing to it. Yesterday I added voice, and it changes the dynamic completely.

The setup: 20 minutes, free tier

The whole thing runs on ElevenLabs for text-to-speech. The free tier is genuinely enough — you get a generous monthly character allowance, plenty for daily conversational use.

Setup steps:

  1. Create a free account at elevenlabs.io
  2. Grab your API key from the profile settings
  3. Add it to your OpenClaw config:
"talk": {
  "provider": "elevenlabs",
  "providers": {
    "elevenlabs": {
      "apiKey": "YOUR_API_KEY_HERE"
    }
  }
}

That’s it. Restart the gateway and voice is live.

How it works

Telegram supports voice messages natively — hold the mic button, speak, send. OpenClaw handles the rest:

  1. Transcription — your voice message is passed through Whisper (OpenAI’s speech model). Fast, accurate, handles accents well.
  2. Processing — the transcribed text reaches the AI exactly as if you’d typed it.
  3. Voice reply — OpenClaw responds via ElevenLabs TTS, sending an audio message back in Telegram.

The full loop: you speak → AI thinks → AI speaks back. No typing at either end. After setting it up yesterday I had my first proper back-and-forth voice conversation with the assistant — it’s a noticeably different experience from text.

When it actually matters

The killer use case is commuting. On the Tube, hands full, bag on one shoulder — the last thing you want is to type a long message on a phone screen. You just want to say it.

A few things that now happen naturally:

  • “What’s on my calendar tomorrow?” — voice message on the way to the station, answer back before you reach the platform.
  • “Add a task to follow up with the bank this week” — faster to say than type, lands straight in Google Tasks.
  • “What was that email about?” — search and summarise while making coffee.
  • Evening wind-down — checking in with the assistant while doing something else, no screen required.

The friction of phone typing is real and constant. Voice removes it for anything conversational.

What works well

  • Whisper accuracy is excellent — even with background noise (train, street) it handles it well
  • Short factual queries are perfect — calendar, tasks, quick lookups
  • Multilingual — Whisper handles Russian just fine if you switch languages mid-conversation

What to watch out for

  • Long responses don’t translate well to audio — if you want a summary or a list, ask for a text reply. Mix and match based on context.
  • Free tier limits — ElevenLabs free tier is generous but not unlimited. For heavy daily use, worth keeping an eye on the dashboard.
  • Quiet environments — TTS replies play as audio in Telegram, which is great on headphones and awkward in a meeting room.

Worth it?

Yes, clearly. Twenty minutes of setup, free tier account, and you get a genuinely hands-free interface to your assistant. It’s not a gimmick — voice is just a better input method for a lot of real situations.

If you’ve already got OpenClaw running with Telegram, this is the easiest upgrade you can make.

Not running OpenClaw yet? Start with the Day 1 setup guide.