Your OpenClaw server. Your voice. No screen required.

Most voice assistants use basic TTS and push-to-talk buttons.

ClawVoice uses Gemini Live API — a real-time bidirectional audio stream. You talk naturally, Gemini listens, understands, and responds with native voice synthesis. You can interrupt it mid-sentence. It hears you even with the screen off. All the heavy lifting — memory, tools, web search, messaging — runs on your own OpenClaw server with whichever AI model you've configured.

• 🎙 Always listening — tap once to start, speak naturally
• 🔒 Screen off — lock your phone and keep talking; audio session stays alive
• 🎧 Any headphones — AirPods, wired, Bluetooth; speaker mode as fallback
• ⚡ Gemini 2.5 Flash — the fastest native audio model available (gemini-2.5-flash-native-audio-preview-12-2025)
• 🤖 OpenClaw backend — your AI model executes actions: web search, messages, reminders, smart home, anything your server can do (supports any provider OpenClaw supports)
• 🍎 Hey Siri activation — say "Hey Siri, [your phrase]" to launch and start listening immediately
• 🔋 Battery friendly — minimal UI, no video, no camera, no polling
• 🔐 Private — your keys, your server, no third-party cloud routing
• 💬 Session history — conversations are saved and accessible from the sessions drawer; all data stays on your OpenClaw server

You speak
    │  PCM audio · 16 kHz
    ▼
Gemini Live API (Google)
    │  understands intent → calls execute(task)
    ▼
ClawVoice app
    │  POST /v1/chat/completions
    ▼
OpenClaw server  (your machine, via Tailscale)
    │  your AI model runs tools — search, messages, calendar, skills...
    ▼
result text → Gemini synthesizes natural voice response
    │  PCM audio · 24 kHz
    ▼
Your headphones or speaker

Session context: each voice session maps to exactly one OpenClaw server session. The context accumulates naturally as you talk. Pause or close the app → next session starts fresh with a clean context.

Privacy: ClawVoice does not store your conversations. All session history — messages, tool calls, responses — is stored exclusively on your own OpenClaw server. Nothing is sent to any third-party service except Google's Gemini Live API for real-time audio processing.

You need all three before setup:

An OpenClaw gateway running on your machine with /v1/chat/completions enabled:

// ~/.openclaw/openclaw.json
{
  "gateway": {
    "port": 18789,
    "http": {
      "endpoints": {
        "chatCompletions": { "enabled": true }
      }
    },
    "auth": {
      "mode": "token",
      "token": "your-gateway-token"
    }
  }
}

openclaw gateway restart

Tailscale connects your iPhone to your home server over an encrypted VPN — no port forwarding, works anywhere, free tier is enough.

Install on your server, then get the machine's Tailscale URL from login.tailscale.com — it looks like https://your-machine.tail06d585.ts.net.

Install on your iPhone from the App Store and log in to the same account.

Same Wi-Fi? You can skip Tailscale and use your server's local IP: http://192.168.1.x:18789.

Get a free key at aistudio.google.com/apikey. The Live API is included in the free tier.

git clone https://github.com/lucasudar/clawvoice-ios.git
open clawvoice-ios/ClawVoice.xcodeproj

1. Signing → Xcode → ClawVoice target → Signing & Capabilities → set your Team and a unique Bundle ID
2. Keys → tap ⚙️ in the app after launch, enter your Gemini key, OpenClaw URL and token
3. Run → Cmd+R → allow microphone when prompted

A free Apple ID is enough to run on your own device — no paid developer account needed.

iOS doesn't allow fully custom always-on wake words, but Siri Shortcuts let you pick any phrase as a trigger:

1. Shortcuts app → + → Add Action → search ClawVoice → Activate Assistant
2. Tap the shortcut name → Add to Siri → record any phrase you like (e.g. your assistant's name, a nickname, anything you want)
3. Say "Hey Siri, [your phrase]" — app launches and immediately starts active listening

No button tap needed. Works with screen off and any headphones paired with your phone.

"Connection Error" on first tap — check your Gemini API key in Settings ⚙️

OpenClaw not responding — confirm Tailscale is connected on both devices; verify chatCompletions.enabled: true in your gateway config

Gemini error 1008 — try a different model in Settings; Google occasionally renames Live API models. Try gemini-2.0-flash-live-001 as fallback

Echo / assistant hears itself — happens when using phone speaker without headphones; the 500ms mute window after each response suppresses most echo. Use headphones for best results

Choppy audio on Bluetooth — prefer AirPods or wired headphones; generic BT headsets sometimes add codec latency

• SwiftUI · iOS 17+
• URLSessionWebSocketTask · no SPM dependencies
• AVAudioEngine · PCM 16 kHz capture, 24 kHz Float32 playback
• Gemini Live API · v1beta WebSocket · gemini-2.5-flash-native-audio-preview-12-2025
• OpenClaw /v1/chat/completions · one server session per voice session (new session on pause or app close)