Double-clap. Jarvis wakes up, greets you with the weather and your tasks, answers your questions with dry British wit, controls your browser, and sees your screen.

Built entirely with Claude Code — no code written manually.

Demo & Explaination

• Double-Clap Trigger — Clap twice and your entire workspace launches: Spotify, VS Code, Obsidian, Chrome with Jarvis UI
• Voice Conversation — Speak freely with Jarvis through your microphone. He listens, thinks, and responds with voice
• Sarcastic British Butler — Jarvis speaks German with the personality of Tony Stark's AI: dry, witty, and always one step ahead
• Weather & Tasks — On startup, Jarvis greets you with the current weather and a humorous summary of your open tasks from Obsidian
• Browser Automation — "Search for MiroFish" → Jarvis opens a real browser, navigates to the page, reads the content, and summarizes it for you
• Screen Vision — "What's on my screen?" → Jarvis takes a screenshot, analyzes it with Claude Vision, and describes what he sees
• World News — "What's happening in the world?" → Jarvis opens worldmonitor.app and summarizes current global events
• Window Snapping — All launched apps automatically snap into quadrants on your screen

You (speak) → Chrome Browser (Web Speech API) → FastAPI Server (local)
                                                       ↓
                                                Claude Haiku (thinks)
                                                       ↓
                                    ┌──────────────────┼───────────────────┐
                                    ↓                  ↓                   ↓
                             ElevenLabs TTS     Playwright Browser    Screen Capture
                             (speaks back)      (searches/opens)     (Claude Vision)
                                    ↓
                             Audio → Chrome → You (hear)

• Windows 10/11
• Python 3.10+
• Google Chrome
• Claude Code (recommended for setup)

1. Clone the repo:

   git clone https://github.com/Julian-Ivanov/jarvis-voice-assistant.git
   cd jarvis-voice-assistant

2. Open in VS Code, start Claude Code, and say:

   Set up Jarvis for me.
   

3. Claude Code will ask for your API keys, name, preferences, and configure everything automatically.

1. Clone and install dependencies:

   git clone https://github.com/Julian-Ivanov/jarvis-voice-assistant.git
   cd jarvis-voice-assistant
   pip install -r requirements.txt
   playwright install chromium

2. Create config.json from the template:

   cp config.example.json config.json

3. Edit config.json with your API keys and preferences:

   {
     "anthropic_api_key": "sk-ant-...",
     "elevenlabs_api_key": "sk_...",
     "elevenlabs_voice_id": "YOUR_VOICE_ID",
     "user_name": "Your Name",
     "user_address": "Sir",
     "city": "Hamburg",
     "workspace_path": "C:\\path\\to\\jarvis-voice-assistant",
     "spotify_track": "spotify:track:YOUR_TRACK_ID",
     "browser_url": "https://your-website.com",
     "obsidian_inbox_path": "C:\\path\\to\\obsidian\\inbox",
     "apps": ["obsidian://open"]
   }

4. Start Jarvis:

   python server.py

5. Open Chrome and go to http://localhost:8340

6. Click anywhere on the page, then speak!

python server.py

Then open http://localhost:8340 in Chrome.

python scripts/clap-trigger.py

Clap twice → Spotify plays your song, VS Code opens, Obsidian opens, Chrome opens with Jarvis. All windows snap into quadrants.

1. Open Task Scheduler (Win + R → taskschd.msc)
2. Create Task → Trigger: "At log on"
3. Action: powershell with argument:

   -ExecutionPolicy Bypass -Command "python C:\path\to\scripts\clap-trigger.py"
   

jarvis-voice-assistant/
├── server.py              # FastAPI backend — the brain
├── browser_tools.py       # Playwright browser automation
├── screen_capture.py      # Screenshot + Claude Vision
├── config.json            # Your personal config (gitignored)
├── config.example.json    # Template for new users
├── requirements.txt       # Python dependencies
├── frontend/
│   ├── index.html         # Jarvis web UI
│   ├── main.js            # Speech recognition + WebSocket + audio
│   └── style.css          # Dark theme with animated orb
├── scripts/
│   ├── clap-trigger.py    # Double-clap detection
│   └── launch-session.ps1 # Launches all apps + window snapping
├── CLAUDE.md              # Instructions for Claude Code
└── SETUP.md               # Detailed setup guide

Edit the system prompt in server.py → build_system_prompt(). The personality, greeting behavior, and action instructions are all defined there.

Edit config.json:

{
  "spotify_track": "spotify:track:YOUR_TRACK_ID",
  "browser_url": "https://your-website.com",
  "apps": ["obsidian://open", "slack://"]
}

Find a voice on elevenlabs.io, copy the Voice ID, and set it in config.json:

{
  "elevenlabs_voice_id": "YOUR_VOICE_ID"
}

{
  "city": "Berlin"
}

In scripts/clap-trigger.py:

THRESHOLD = 0.15  # Lower = more sensitive
MAX_GAP = 1.2     # Seconds between claps

This template is built for Windows. If you're on macOS, clone the repo and tell Claude Code:

Convert this project to work on macOS.

Claude Code will adapt the PowerShell scripts to shell scripts, adjust paths, and handle macOS-specific differences.

• FastAPI — Python web framework for the local server
• Claude Haiku — Fast, affordable AI model (the brain)
• ElevenLabs — Natural text-to-speech (the voice)
• Playwright — Browser automation
• Web Speech API — Browser-native speech recognition
• sounddevice — Audio input for clap detection

Built by Julian with Claude Code.

Inspired by Iron Man's J.A.R.V.I.S. — "At your service, Sir."

MIT — use it, modify it, build on it. If you build something cool, let me know!