<div align="center"> <img src="backend/src/ui/assets/vebgen_hero_banner.png" alt="VebGen - The World's First AST-Powered AI Development Agent" width="100%"> </div> <br> <!-- Your existing content starts below --> <div align="center">

<img src="backend/src/ui/assets/vebgen_logo.png" width="48" alt="VebGen Logo" /> VebGen

🏆 The World's First AST-Powered AI Development Agent

Your AI Development Team That Never Sleeps

Zero-token code intelligence • Military-grade sandbox • Dual-agent self-remediation

🌐 Website: vebgen.dev (coming soon)

✨ Features • 🆚 Why Different? • 🚀 Quick Start • 📚 Docs Hub • Discord • 🗺️ Roadmap • 🤝 Contributing

</div> <div align="center">

📊 Quick Stats

📦 900KB Code • ✅ 356 Tests • 📚 1,000KB Docs • 🎨 WCAG 2.1 Built-In • 💰 $0 Spent • ⏱️ 8 Months • 👥 2-Person Team • Made in India 🇮🇳

</div>

---

🎯 What is VebGen?

VebGen is the first AI-powered development agent that understands your code without reading it.

While tools like Cursor and GitHub Copilot help you write code faster, VebGen builds complete, production-ready applications from natural language—with innovations no other agent has:

📢 Early Feedback:

"VebGen is the future of AI development - I built a full Django REST API in 10 minutes!" - Early Adopter

"Finally, an AI agent that doesn't break my code!" - Beta Tester

Major Additions

| Feature | What It Does | Impact | |---------|--------------|--------| | 🎨 Frontend Validation Suite | 7 specialized parsers/analyzers enforce WCAG 2.1, detect render-blocking scripts, unused CSS, layout thrashing | ✅ Production-ready accessibility by default<br>✅ No manual Lighthouse audits needed<br>✅ 47 new frontend tests | | 📂 External Project Support | Load ANY Django project—not just VebGen-created ones—and continue building | ✅ Adopt existing codebases<br>✅ Migrate from Cursor/Copilot mid-project<br>✅ Work on open source projects | | 🔧 Search/Replace Patching | Surgical code edits with fuzzy matching (80%+ similarity) when strict patches fail | ✅ 70% → 92% patch success rate<br>✅ Fewer TARS remediation loops | | 🛡️ State Corruption Auto-Recovery | 3-tier recovery (backups → rebuild from code) when project state is invalid | ✅ Never lose project history<br>✅ Automatic disk corruption handling |

📖 Full v0.3.0 Release Notes: releases/v0.3.0.md
📖 v0.2.0 Historical Release: releases/v0.2.0.md

---

🏆 World Firsts

| Innovation | What It Means | Why It Matters | |------------|---------------|----------------| | 🧠 AST-Powered Code Intelligence | Understands Django/Flask projects by parsing code structure (95+ constructs), not dumping files to LLM | ✅ 10x faster context building<br>✅ Zero wasted tokens<br>✅ Understands relationships between models, views, URLs | | 🔒 Built-In Sandbox Security | Military-grade file system isolation and command whitelisting—no Docker required | ✅ Works natively on Windows/Mac/Linux<br>✅ Blocks path traversal, shell injection<br>✅ 50+ safe commands whitelisted | | 🔄 Dual-Agent Self-Remediation | When tests fail, Tars analyzes the error and Case patches the bug—autonomously | ✅ Fixes 70% of errors without human input<br>✅ Learns from mistakes<br>✅ Up to 3 remediation strategies | | 💰 100% Free-Tier Built | Entire 900KB codebase, 356 tests, 1,000KB docs built without spending $1 on APIs | ✅ Proves free-tier APIs (Gemini, OpenRouter) can build world-class software<br>✅ No venture capital required | | 🎨 Frontend Quality Enforcement (v0.3.0) | Automated WCAG 2.1 + Lighthouse audits block feature completion until accessibility/performance issues are fixed | ✅ Production-ready UI by default<br>✅ No manual audits needed<br>✅ 12+ WCAG criteria covered |

---

💎 How is VebGen Different from Copilot or Cursor?

While tools like GitHub Copilot and Cursor are amazing for helping you write code faster, VebGen works on a whole different level. Think of it less like a code-completion assistant and more like hiring an entire development team to build a project for you—from scratch to deployment.

📊 VebGen vs. The Competition

| Feature | VebGen | Cursor | GitHub Copilot | Devin | AgentScript | |---------|--------|--------|---------------|-------|-------------| | Code Intelligence | ✅ AST-based (zero-token) | ❌ Full-file context | ❌ Full-file context | ⚠️ Cloud-based | ⚠️ Limited | | Frontend Validation | ✅ WCAG 2.1 + Lighthouse (built-in) | ❌ None | ❌ None | ⚠️ Manual audits | ❌ None | | Sandbox Security | ✅ Built-in (no Docker) | ❌ None | ❌ None | ✅ Cloud sandbox | ⚠️ AST interpreter only | | Self-Remediation | ✅ Dual-agent (Tars+Case) | ⚠️ Retry only | ❌ None | ✅ Single-agent | ❌ None | | Full-App Generation | ✅ Django, Flask, React | ❌ Code completion | ❌ Code completion | ✅ Full-stack | ⚠️ Limited | | Desktop App | ✅ Native Windows/Mac/Linux | ❌ VS Code only | ❌ IDE plugin | ❌ Cloud only | ❌ CLI only | | Cost | ✅ Free (BYOK) | $20/month | $10/month | $500/month | Free (BYOK) | | Open Source | ✅ MIT License | ❌ Proprietary | ❌ Proprietary | ❌ Proprietary | ✅ Open |

TL;DR: VebGen is the only agent with AST-based intelligence, built-in sandboxing, AND dual-agent remediation—all for free. 🚀

The Game-Changing Differences

🧬 Zero-Token AST Parsing

VebGen reads your entire codebase without consuming any API tokens. While Cursor charges you to analyze your own code, VebGen uses Abstract Syntax Trees (AST) to parse Python/Django projects locally.

💡 What this means: Cursor and Copilot charge you (in tokens) to read your own code. VebGen never does. Your existing codebase analysis is 100% free, 100% instant.

📖 Technical Deep Dive: See Code Intelligence Service Documentation for AST parsing implementation details.

🛡️ Military-Grade Command Security

Afraid of AI agents accidentally running rm -rf / or exposing your secrets? VebGen has production-ready security validation that never trusts the LLM:

• ✅ Whitelist-only execution (only approved commands run: pip, git, django-admin)
• ✅ Regex blocklist (dangerous commands auto-blocked with safe alternatives suggested)
• ✅ Multi-layer path validation (prevents any file access outside your project)
• ✅ User confirmation required for all sensitive operations
• ✅ No direct shell access for the AI—ever

💡 What this means: Other tools let the LLM execute commands with basic validation. VebGen treats every command like a security audit. Your production server stays 100% safe.

📖 Security Implementation: See Command Executor Documentation for whitelist/blocklist mechanisms and Secure Storage Documentation for API key encryption.

💾 Never Lose Progress—Ever

VebGen's Memory Manager ensures you can close the app anytime and continue exactly where you left off—even if something goes wrong.

Here's what actually happens behind the scenes:

📂 Automatic State Persistence Every action (files created, commands run, features completed) is saved to a hidden .vebgen/ folder in your project. Close VebGen mid-development? Reopen it, and you're instantly back in the same state—no need to explain context again.

💡 What this means: Unlike chat-based tools that lose context when you close the tab, VebGen remembers everything. The AI picks up exactly where it stopped—with full memory of what's been built, what's pending, and what failed.

📖 State Management Details: See Memory Manager Documentation for backup algorithms and Project Models Documentation for state schema.

🔐 Corruption Protection with SHA-256 Verification
Every time VebGen saves your project state, it calculates a SHA-256 cryptographic hash of the data. When loading, it verifies the hash matches—if it doesn't (file corruption, system crash, disk error), VebGen automatically restores from the most recent valid backup.

🗂️ 5 Automatic Backups (Timestamped)
Before saving any state changes, VebGen creates a timestamped backup. It keeps the 5 most recent backups, pruning older ones automatically. If all 5 backups are