Open JARVIS — Local-First AI Operating System

A fully modular, cinematic, local-first AI assistant inspired by Iron Man's JARVIS. Features a stunning 3D Arc Reactor visualization, real-time tool execution, persistent memory, voice interaction, and a Tauri desktop app — all running locally on your machine.

Features

• JARVIS Personality — British-butler AI that calls you "sir", created by Aniket
• Local LLM — Runs on Ollama with custom jarvis model (qwen2.5:1.5b), falls back to Gemini API
• 8 Built-in Tools — System info, weather, web search, datetime, terminal, file read/write, screenshot
• Persistent Memory — Conversation history saved across sessions (50 turns)
• 3D Arc Reactor Engine — Three.js visualization with breathing animation, bloom, orbital rings
• Voice I/O — Browser-native speech recognition + text-to-speech with JARVIS voice
• Streaming Chat — Real-time SSE token streaming with markdown stripping for TTS
• Smart Routing — 4-tier LLM routing: intent classification → complexity scoring → model selection
• Desktop App — Tauri 2 native app with system tray, global hotkeys, auto-backend launch

Architecture

src/
├── frontend/          React 18 + Three.js + R3F + Vite (port 1420)
├── backend/           Python FastAPI + Agents + Tools (port 8420)
│   ├── agents/        Brain orchestrator + 8 tool agents
│   ├── llm/           4-provider routing (Ollama, Gemini, OpenRouter, GGUF)
│   ├── memory/        Persistent conversation history
│   └── api/           REST + WebSocket endpoints
├── desktop/           Tauri 2.x Desktop Shell (Rust)
└── shared/            Shared types & constants

docs/
└── architecture/      Design documents

Tech Stack

Layer	Technology
Desktop	Tauri 2.x (Rust) with system tray & hotkeys
Frontend	React 18, Three.js/R3F, GSAP, Framer Motion, TailwindCSS
Backend	Python FastAPI, WebSockets, asyncio
AI/LLM	Ollama (local), Gemini API, OpenRouter
Voice	Web Speech API (STT + TTS)
State	Zustand (8 stores)

Quick Start

Prerequisites

• Node.js 20+
• Rust & Cargo (for Tauri)
• Python 3.11+
• Ollama (for local LLMs)

Setup

# Frontend
cd src/frontend && npm install

# Backend
cd src/backend && pip install -e .

# Desktop (Tauri)
cd src/desktop && cargo build

# Start development
npm run dev:frontend    # Port 1420
npm run backend         # Port 8420
npm run dev:desktop     # Tauri shell

Hardware Tiers

Tier	RAM	GPU VRAM	Recommended Model
Minimal	8 GB	None	Phi-3 Mini 3.8B Q4
Low	10 GB	2 GB	Llama 3.2 3B Q4
Medium	16 GB	6 GB	Llama 3.1 8B Q6
High	24 GB	8 GB	Llama 3.1 8B F16
Ultra	32 GB	12 GB+	Llama 3.1 70B Q4

License

MIT