Always-on, tool-using AI assistant activated by double-clap detection.

Jarvis is a persistent desktop AI agent that listens for a double-clap, takes voice commands, executes actions on your computer, and speaks results back. It uses Gemini 2.0 Flash with native function calling to search the web, control your browser, open apps, manage focus sessions, and more.

JARVIS DAEMON (always-on)

  ACTIVATION ──> EARS ──> BRAIN ──> HANDS (tools)
  (clap detect)  (STT)    (Gemini)   |
                                      ├── web_search
       VOICE <──────────────────────  ├── open_browser_tabs
       (TTS)                          ├── open_application
                                      ├── close_browser_tab
       FACE <───────────────────────  ├── system_command
       (tray + HUD)                   ├── focus_mode
                                      ├── clipboard
                                      └── set_reminder

1. Download the latest release from GitHub Releases
2. Extract the zip file
3. Double-click setup.bat
4. Follow the prompts — you'll need a Gemini API key (free)

That's it. Jarvis will auto-start on login and run in the background.

The Windows bootstrapper creates a local virtual environment, stops a previous Jarvis install if one is running, repairs locked installs by falling back to a fresh side-by-side environment, and launches the visual setup wizard. The wizard saves the startup sequence defaults for the first launch: project directory, browser URLs, app defaults, Warp, and Claude settings.

Manual Windows launcher: %LOCALAPPDATA%\JarvisAI\Jarvis AI.cmd

git clone https://github.com/minutecoder34-cmd/JarvisAi.git
cd JarvisAi/JarvisAi-2.0.0

# Install the package
pip install .

# Run the setup wizard (creates config + auto-start)
jarvis install

# Start Jarvis
jarvis start

pip install -e ".[dev]"

You need a Gemini API key (required) and optionally a Google CSE key for web search.

The installer will prompt for these. You can also edit ~/.jarvis/.env directly:

GOOGLE_API_KEY=your_gemini_api_key
SEARCH_API_KEY=your_google_cse_or_serpapi_key
SEARCH_ENGINE_ID=your_google_cse_id
# ELEVENLABS_API_KEY=optional_premium_tts

jarvis start                  # Start daemon (background with HUD/tray)
jarvis start --headless       # No GUI, lower RAM
jarvis start --test           # Text mode (stdin, no mic)
jarvis stop                   # Stop the daemon
jarvis status                 # Check daemon status
jarvis text "open spotify"    # Send a text command
jarvis calibrate              # Re-calibrate clap detection
jarvis config                 # Edit configuration
jarvis log                    # Tail today's conversation log
jarvis uninstall              # Remove auto-start

1. Double-clap performs the initial Jarvis startup sequence
2. Jarvis plays the greeting/opening flow and launches the configured apps/tabs
3. Wake word "Jarvis" arms for follow-up voice commands after initialization
4. Faster-Whisper transcribes your voice command locally
5. Gemini 2.0 Flash decides which tools to call via function calling
6. Tools execute: search the web, open tabs/apps, control system
7. Jarvis speaks a concise summary and opens sources in your browser

jarvis/
├── shared/          # Shared types, events, config
├── activation/      # Double-clap detection + mic manager
├── ears/            # Speech-to-text (faster-whisper)
├── brain/           # Gemini orchestrator + tool definitions
├── hands/           # Tool executor + platform abstraction
│   └── tools/       # web_search, browser, apps, focus_mode, etc.
├── voice/           # TTS engine + speech queue
├── face/            # System tray + overlay HUD
├── daemon/          # Background service + CLI + installer
└── tests/           # Unit + integration tests

Config file: ~/.jarvis/config.yaml

gemini_model: gemini-2.0-flash
whisper_model_size: base.en    # tiny.en, small.en, medium.en
tts_engine: pyttsx3            # macos_say, elevenlabs, pyttsx3
clap_sensitivity: 0.7
search_provider: google_cse    # google_cse, serpapi
focus_check_interval_s: 30
headless: false
require_initialization_clap: true
startup_project_dir: C:/Users/you/source/project
startup_browser: chrome
startup_urls:
  - https://mail.google.com/
  - https://classroom.google.com/
  - https://docs.google.com/
  - https://www.youtube.com/watch?v=fPO76Jlnz6c&autoplay=1
startup_apps:
  - Obsidian
  - Warp
launch_warp_with_claude: true
claude_command: claude
auto_detect_microphone: true

Windows: Run Uninstall-Jarvis.ps1 or:

jarvis uninstall

This removes the Task Scheduler entry. To fully remove:

• Delete %LOCALAPPDATA%\JarvisAI (the venv)
• Delete ~/.jarvis (config + logs)

# Run all tests
cd jarvis && python -m pytest tests/ -v

# Run specific test suites
python -m pytest tests/test_activation.py -v
python -m pytest tests/test_hands.py -v
python -m pytest tests/test_brain.py -v
python -m pytest tests/test_integration.py -v

Push a version tag to trigger the GitHub Actions release workflow:

git tag v2.0.0
git push origin v2.0.0

This creates a GitHub Release with a downloadable zip.

• LLM: Google Gemini 2.0 Flash (function calling)
• STT: Faster-Whisper (local, int8 quantization)
• TTS: macOS say / ElevenLabs / pyttsx3
• Audio: sounddevice, scipy, webrtcvad
• GUI: PyQt6 + qasync
• Search: Google Custom Search / SerpAPI
• Daemon: launchd / systemd / Task Scheduler