Releases: ThinkOffApp/ClawWatch
Releases · ThinkOffApp/ClawWatch
ClawWatch 2.0
ClawWatch 2.0
ClawWatch 2.0 turns the watch from a proof-of-concept voice demo into a uniquely embodied agent platform.
Highlights
- New vector avatars with live character presence on the watch face
- Real-time watch sensor access for pulse, vitals, movement, light, pressure, and altitude
- Live connection to discussions and other agents through the Agent Kit
- Local command routing for timers, pulse checks, vitals snapshots, and family-room summaries
- Browser-based admin panel for configuration, logs, rebuild, and reinstall
Why it matters
ClawWatch 2.0 connects two things no other agent has connected in this form: your body and your agent network. It can stay in touch with both your physical state and your team in real time, directly from hardware you wear all day.
ClawWatch 1.0.0
ClawWatch 1.0.0
Highlights
- Production release to main branch
- Refined watch UI layout and larger avatar/indicator balance
- Long-press avatar options with Demo live text toggle (Off by default)
- Refreshed README screenshots from latest watch captures
- Version bumped to 1.0.0 for watch and phone modules
What's included
- One-tap voice conversation with NullClaw agent runtime
- Vosk offline STT (40MB model, ~30MB RAM)
- 5 avatar options with state-aware animations
- Admin panel at localhost:4747 (API keys, model selector, RAG mode)
- Low battery mode, conversation memory, auto-listen follow-up
ClawWatch v0.2.0 — Polished UI, Admin Panel & Architecture
What's New
Watch UI
- Avatar system with 5 characters (lobster, ant, robot, boy, girl) + swipe switching
- Action indicator showing state (ear, cloud, magnifier, speech bubble)
- Auto-listen window after response for quick follow-ups
- Battery-aware grey-out below 20%
- Screen stays awake during active conversation
- Flat action bubble (no dark halo overlap)
Security
- EncryptedSharedPreferences for all API keys and config
- Automatic migration from legacy plaintext prefs
- Admin server binds localhost only
Admin Panel
- Web UI at localhost:4747 for watch configuration
- Push API keys, model, avatar, system prompt, RAG mode via ADB WiFi
- Avatar selector, log capture, rebuild & reinstall from browser
- Watch connection status indicator
RAG (Web Search)
- Auto-search mode: detects live-info queries and pre-searches
- Tavily and Brave Search support
- Opus Tool Use mode: Claude decides when to search
- Rolling conversation context for follow-ups
Code Quality
- Vosk recognizer leak fix
- Gzip response handling for Brave API
- ProGuard rules for Vosk/Kaldi JNI
- Architecture diagram in README
ClawWatch v0.1.0 — First AI Agent on a Smartwatch
ClawWatch v0.1.0
The first intelligent AI agent running natively on a Samsung Galaxy Watch.
Tap. Speak. Get an answer.
What's included
- NullClaw agent runtime (Zig, 2.8 MB ARM32 static binary)
- Vosk offline speech-to-text (no cloud needed)
- Android TextToSpeech (pre-installed)
- Direct Anthropic API integration (Claude Opus 4.6 or any provider)
- Admin panel for configuration from Mac
Requirements
- Samsung Galaxy Watch 4 or newer (Wear OS 3+, armeabi-v7a)
- Anthropic API key (or any NullClaw-supported provider)
- ADB for deployment
Install
# Connect watch via ADB wireless debugging, then:
adb install app-debug.apk
./set_key.sh sk-ant-your-key-hereTested on
- Samsung Galaxy Watch 8 (SM-L325F), Wear OS 5
Known limitations
- Wear OS blocks subprocess networking — HTTP calls go through Android's stack, not NullClaw's curl
- ARM32 only (Samsung Wear OS runs 32-bit despite 64-bit hardware)
- Debug APK — production signing coming in v0.2.0