Skip to content

Interactive 3D codebase visualization using AST parsing, semantic embeddings, and graph analysis. Built with WebAssembly, WebGPU, and modern web APIs.

Notifications You must be signed in to change notification settings

yuan-cloud/helios

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

HELIOS 🌌

See the gravity of your code — Interactive 3D visualization of codebase structure using AST parsing, semantic embeddings, and graph analysis.

Live Demo License

HELIOS is a client-side only web application that visualizes codebases as interactive 3D force-directed graphs. It runs entirely in your browser—no server, no data uploads, complete privacy.

✨ Features

🎯 Core Capabilities

  • 🌳 AST Parsing: Language-aware parsing using Tree-sitter (WASM) for JavaScript/TypeScript and Python
  • 🔗 Call Graph Extraction: Static analysis to identify function calls and dependencies
  • 🧠 Semantic Similarity: AI-powered embeddings (Transformers.js) to find semantically related functions
  • 📊 Graph Analysis: Centrality metrics, community detection (Louvain), and network analysis
  • 💾 Persistent Storage: SQLite-WASM with OPFS for local data persistence
  • 🎨 3D Visualization: Interactive force-directed graph with Three.js and 3d-force-graph
  • ⚡ WebGPU Acceleration: Automatic WebGPU detection for faster embeddings (with WASM fallback)

🚀 Technical Highlights

  • Zero Backend: Everything runs client-side—perfect for privacy-sensitive codebases
  • WebAssembly: Tree-sitter, SQLite, and ONNX Runtime compiled to WASM
  • Cross-Origin Isolation: Service worker enables SharedArrayBuffer and WASM threads
  • Modern Web APIs: File System Access API, OPFS, WebGPU, Web Workers
  • No Build Step: Pure ES modules with import maps—deploy as-is

🎬 Quick Start

Try the Demo

  1. Visit the live demo
  2. Click "Load Demo" to see a sample codebase visualization
  3. Or "Select Repository" to analyze your own code

Local Development

# Clone the repository
git clone https://github.com/yuan-cloud/helios.git
cd helios

# Install dependencies (optional - for local vendor files)
npm install

# Serve locally
python3 -m http.server 8000
# Visit http://localhost:8000

Note: For full functionality (OPFS, WebGPU), serve over HTTPS or use localhost.

📖 How It Works

  1. Select Repository: Choose a local folder using the File System Access API
  2. Parse: Tree-sitter extracts functions, imports, and call sites from your code
  3. Embed: Transformers.js generates semantic embeddings for each function
  4. Analyze: Graphology computes network metrics (centrality, communities)
  5. Visualize: Interactive 3D graph shows code structure and relationships

Architecture

┌─────────────────┐
│   User Selects  │
│   Repository    │
└────────┬────────┘
         │
    ┌────▼─────┐
    │  Parser  │ → Functions, Calls, Imports
    │ (WASM)   │
    └────┬─────┘
         │
    ┌────▼────────┐
    │ Embeddings  │ → Semantic Vectors
    │ (WebGPU)    │
    └────┬────────┘
         │
    ┌────▼─────┐
    │  Graph   │ → Centrality, Communities
    │ Analysis │
    └────┬─────┘
         │
    ┌────▼──────────┐
    │ Visualization │ → 3D Interactive Graph
    │  (Three.js)   │
    └───────────────┘

🛠️ Tech Stack

Component Technology
Parsing web-tree-sitter (WASM)
Embeddings Transformers.js + ONNX Runtime Web
Graph Analysis Graphology + Louvain communities
Storage SQLite-WASM + OPFS
Visualization Three.js + 3d-force-graph
UI Framework Alpine.js + Tailwind CSS
Hosting Static (Cloudflare Pages / GitHub Pages)

📚 Documentation

🎯 Use Cases

  • Code Exploration: Understand large codebases through visual navigation
  • Refactoring Planning: Identify tightly coupled modules and dependencies
  • Onboarding: Help new team members understand codebase structure
  • Architecture Review: Visualize system design and identify patterns
  • Research: Analyze code organization and semantic relationships

🔒 Privacy & Security

  • 100% Client-Side: No code is uploaded to any server
  • Local Storage: All data stored in browser (OPFS)
  • No Analytics: Zero tracking or telemetry
  • Open Source: Full source code available for audit

🌟 Key Differentiators

  1. Privacy-First: Unlike cloud-based tools, HELIOS never sends your code anywhere
  2. No Build Required: Pure ES modules—works immediately after deployment
  3. Modern Web Standards: Leverages latest browser APIs (WebGPU, OPFS, SharedArrayBuffer)
  4. Extensible: Easy to add new languages via Tree-sitter grammars
  5. Performance: WebGPU acceleration for embeddings, WASM for parsing

🚧 Current Status

MVP Complete

  • AST parsing (JS/TS, Python)
  • Call graph extraction
  • Semantic embeddings
  • Graph analysis
  • 3D visualization
  • OPFS persistence
  • WebGPU support

Roadmap 🗺️

  • Additional languages (Go, Rust, Java)
  • Stack graphs for better name resolution
  • HNSW approximate nearest neighbor search
  • UMAP layout seeding
  • Project snapshot export/import

🤝 Contributing

This is a personal portfolio project. While I appreciate interest in HELIOS, I'm not currently accepting contributions or pull requests. This project serves as a showcase of my technical capabilities.

If you have questions about the implementation or want to discuss the technology, feel free to open a GitHub Discussion or reach out!

📄 License

MIT License - see LICENSE file for details.

🙏 Acknowledgments

MCP Agent Mail

HELIOS was built using MCP Agent Mail — a coordination system that enabled 5 AI agents to work together seamlessly.

Created by: @Dicklesworthstone

Without MCP Agent Mail, coordinating multiple AI agents would have been impossible. The system provided file reservations, message coordination, and a complete audit trail — enabling 341 commits with zero merge conflicts.

Learn more: See docs/MCP_AGENT_MAIL_SHARING.md for details on how it works.

Libraries & Tools

Built with:


Made with ❤️ for developers who want to see their code in a new dimension.