README.md

🚀 NeuralForge

A Lightweight AI Code Editor - Tauri-based IDE with local AI, built for privacy and performance.

⚠️ Early Development Stage - Phase 1-2 Complete (Backend), Phase 3 Starting (IDE)

Roadmap • Architecture • Contributing

---

🔥 Why NeuralForge?

Tired of sending your code to the cloud? Frustrated with 200MB+ RAM IDEs? Want an AI that actually learns YOUR coding style?

NeuralForge is a lightweight Tauri-based IDE with embedded AI models that:

• 🧠 Learns from your code - Adapts to your style over time
• 🔒 100% local - Your code never leaves your machine
• ⚡ ~300MB RAM - 75% less than Electron-based editors
• 💬 AI Chat with file manipulation - Direct project changes from chat
• 🆓 Forever free - No subscriptions, no API keys required (optional providers available)
• 🚀 <50ms inference - Faster than cloud solutions

🎯 Key Characteristics

Feature	Description
Tauri-Based	Native performance, 50MB binary vs 200MB Electron
Monaco Editor	VS Code's editor component (standalone)
Local-First AI	All processing happens on your machine
Privacy	Your code never leaves your computer
Open Source	Fully transparent and community-driven
Multi-Provider Chat	Optional OpenAI/Claude/Gemini support
Learning System	Adapts to your coding style automatically
RAG System	Framework documentation at your fingertips

🎬 Demo

See NeuralForge learning your coding patterns in real-time

✨ Features (v0.1.0 Target)

🎨 Modern IDE

• Monaco Editor: VS Code's editor component with syntax highlighting
• File Explorer: Tree view with folder navigation
• Tab Management: Multi-file editing with persistent state
• Themes: Dark/Light modes with customization

🤖 AI Completion (Ghost Text)

• Inline Suggestions: Gray text preview as you type
• YOLO Mode: Auto-accept completions (toggle on/off)
• Smart Debouncing: 500ms delay, no request spam
• Tab/Esc Handling: Accept or reject suggestions

� AI Chat Panel

• Project-Aware Chat: Knows about your open files
• Multi-Provider Support:
  • Local (default, no API key)
  • OpenAI (optional, GPT-4)
  • Claude (optional, Claude 3.5 Sonnet)
  • Gemini (optional, Gemini 1.5 Pro)
  • Custom Endpoints (your own API)
• Code Actions: Create/modify/delete files directly from chat
• Syntax Highlighting: Code blocks with copy buttons

📚 Learning System

• Style Analyzer: Detects your naming, formatting, comment style
• Pattern Recognition: Learns libraries, test patterns, project structure
• Adaptive Suggestions: AI output matches YOUR preferences
• Learning Dashboard: See what the AI learned about your style

🔍 RAG System (Framework Docs)

• Qdrant Vector DB: Embedded, no external services
• Auto-Detection: Finds Spring Boot, React, etc. from pom.xml/package.json
• Semantic Search: "How to use @Transactional?" finds relevant docs
• Code Snippets: Copy-paste ready examples from official docs

🎯 Vibe Coding

• Context-Aware: AI understands errors, cursor position, selection
• Multi-Step Tasks: "Refactor this class to use dependency injection"
• Progress Tracking: See AI's step-by-step execution plan
• Auto-Fix: AI suggests fixes for compile errors

⚡ Performance

• Startup: <3 seconds (Tauri is fast!)
• Inference: <50ms (Java backend, ONNX Runtime)
• RAM Usage: ~300MB idle (75% less than Electron)
• Binary Size: ~50MB (vs 200MB+ Electron apps)

🚀 Quick Start

⚠️ Not Yet Released

NeuralForge is in early development (Phase 3 starting). Pre-built binaries will be available with v0.1.0.

Current Status:

• ✅ Phase 1-2: Java backend complete (86 tests passing)
• 🔄 Phase 3: Tauri IDE development starting (see ROADMAP.md)

Build from Source (Developers Only)

Prerequisites:

• Node.js 18+ & npm
• Rust 1.70+ (for Tauri)
• Java 21+ & Gradle (for backend)
• Python 3.10+ (for tokenizer)

# Clone repository
git clone https://github.com/ymcbzrgn/neuralforge.git
cd neuralforge

# Backend (already complete, just test it)
cd backend
./gradlew test  # Should show 86 tests passing
cd ..

# Frontend (Phase 3 - under development)
# TODO: Tauri project will be created in Sprint 1.1
# cd neuralforge-ide
# npm install
# npm run tauri dev

Download Models:

cd models
python download_model.py  # Downloads CodeT5+ 220M (~1GB)

💻 System Requirements

Minimum

• OS: Windows 10 / macOS 11 / Ubuntu 20.04
• RAM: 2GB (lightweight!)
• Storage: 5GB (models + app)
• CPU: 2 cores @ 2.0GHz
• GPU: Not required

Recommended

• RAM: 4GB
• Storage: 10GB
• CPU: 4 cores @ 2.5GHz
• GPU: Optional (for faster inference, but not needed)

🗺️ Roadmap

Current Phase: Phase 3 - Tauri IDE Development (6 weeks, 9 sprints)

See detailed ROADMAP.md for:

• ✅ Phase 1-2: Backend Complete (86 tests passing)
• 🔄 Phase 3: IDE Development (Sprint 1-9 detailed)
• 📋 Sprint breakdown with tasks, files, tests

Next Milestone: v0.1.0 (Tauri app + AI completion + chat + learning + RAG)

🤝 Contributing

We love contributions! NeuralForge is built by developers, for developers.

Ways to Contribute

1. Code Contributions

   git checkout -b feature/amazing-feature
   # Make your changes
   git commit -m "Add amazing feature"
   git push origin feature/amazing-feature
   # Open a Pull Request

2. Create Adapters

   • Train adapters for specific frameworks
   • Share them with the community
   • Get recognition in our Hall of Fame

3. Report Issues

   • Bug reports
   • Feature requests
   • Performance improvements

4. Documentation

   • Improve docs
   • Write tutorials
   • Create videos

See CONTRIBUTING.md for details.

📚 Documentation

• ROADMAP.md - Development roadmap and current status
• ARCHITECTURE.md - System architecture (Tauri + Java backend)
• NeuralForge_PRD.md - Product requirements and vision
• CONTRIBUTING.md - How to contribute
• TRAINING.md - LoRA adapter training guide
• SECURITY.md - Security architecture and threat model

🏆 Community

Get Involved

For questions, discussions, and to connect with other contributors:

• Check CONTRIBUTING.md for how to get involved
• Review CLAUDE.md for architecture details
• See ARCHITECTURE.md for technical overview

📊 Benchmarks

Performance benchmarks will be published once the project reaches a stable release with reproducible test cases.

🔐 Security

• ✅ All models run locally
• ✅ No telemetry or analytics
• ✅ Adapter signature verification
• ✅ Sandboxed execution
• ✅ Regular security audits

Found a security issue? Please email security@neuralforge.dev

📜 License

NeuralForge is licensed under the Apache License 2.0 - see LICENSE file.

Third-party Licenses

• VS Code OSS: MIT License
• Models: Apache 2.0 / MIT only
• All dependencies: Compatible licenses

💖 Support

If you find NeuralForge useful, consider:

• ⭐ Starring the repository
• 📝 Contributing code or documentation
• 🐛 Reporting bugs and suggesting features

🌟 Star History

🎯 Philosophy

"AI should amplify human creativity, not exploit it. Your code is your intellectual property. Your patterns are your expertise. Your privacy is non-negotiable."

NeuralForge is built on five principles:

1. Privacy First - Your code never leaves your machine (100% local by default)
2. Lightweight - Fast startup, low RAM usage (Tauri beats Electron)
3. Community Driven - Built by developers, for developers
4. Continuous Learning - Gets better with every line you write
5. Choice - Optional cloud providers (OpenAI/Claude/Gemini) without forcing them

---

Ready to build the future of local-first AI coding?

👷 Contribute to NeuralForge

⚠️ Pre-Alpha Development • Contributions Welcome • Phase 3 Starting

Made with ❤️ by developers who believe in privacy and local-first AI

GitHub • Documentation • Report Bug

---

_{Built by the open-source community. Not another corporate AI tool.}