JARVIS is a voice-first AI assistant for macOS, inspired by the MCU's JARVIS, designed to allow users to interact with their computers naturally via voice. It targets macOS users, developers, and power users seeking a more integrated and conversational computing experience, offering benefits like software building, task management, and information retrieval through spoken commands.

How It Works

The system processes voice input through Chrome's Web Speech API, transmitting it via WebSocket to a FastAPI backend. This backend leverages Anthropic's Claude models (Haiku for speed, Opus for deep tasks) to understand intent and generate responses. Speech synthesis is handled by Fish Audio with a distinct JARVIS voice, streamed back to the user. Core macOS integrations (Calendar, Mail, Notes) are managed via AppleScript, while development tasks are spawned using Claude Code, and web browsing is automated with Playwright.

Quick Start & Requirements

The fastest setup involves cloning the repository and using the claude Claude Code CLI for a guided, step-by-step installation that handles API keys and dependencies. Alternatively, manual setup requires Python 3.11+, Node.js 18+, Google Chrome, an Anthropic API key, a Fish Audio API key, and the Claude Code CLI. Prerequisites include macOS.

Highlighted Details

• Action System: Utilizes action tags (e.g., [ACTION:BUILD], [ACTION:BROWSE], [ACTION:RESEARCH]) to trigger specific system functions and AI-driven tasks.
• Memory System: Employs SQLite with FTS5 for persistent storage of user preferences, decisions, and facts across sessions.
• macOS Native Integration: Leverages AppleScript for direct access to Calendar, Mail (read-only), and Notes, bypassing complex OAuth flows.
• AI Capabilities: Supports both low-latency responses via Claude Haiku and in-depth processing for complex research using Claude Opus.
• Visual Feedback: Features an audio-reactive Three.js particle orb that pulses with JARVIS's voice.

Maintenance & Community

This project is built by Ethan and powered by Anthropic Claude and Fish Audio. The README outlines areas for contribution, such as Linux/Windows support, alternative TTS engines, and alternative LLMs, but does not provide direct links to community channels like Discord or Slack.

Licensing & Compatibility

The software is free for personal, non-commercial use. Commercial use requires obtaining a separate license from ethanplus.ai. Compatibility is limited to macOS due to its reliance on AppleScript for system integrations.

Limitations & Caveats

The primary limitation is its macOS-only platform support, stemming from its reliance on AppleScript. Linux and Windows compatibility are noted as areas for future development. The project actively seeks contributions, suggesting it may still be evolving and potentially not feature-complete or fully stable for all advanced use cases.