Show HN: VebGen – AI agent that understands projects with AST, not LLM tokens

Hey HN! Ramesh here, 20, from India.

Just released VebGen v0.3.0 after 8 months of solo work. Not a single dollar spent - entirely free-tier APIs.

THE PROBLEM:

Tools like Cursor read your codebase by sending files to LLMs. Every. Single. Time. This burns tokens and costs money.

THE SOLUTION:

VebGen uses AST (Abstract Syntax Tree) parsing locally instead. Reads your Django project in 0.2 seconds - all 95+ constructs (models, views, serializers, signals) - without any API calls.

Only when GENERATING code does it use the LLM.

Think: Reading code = AST (free, instant) | Writing code = LLM (paid, smart)

v0.3.0 NEW FEATURES:

• WCAG 2.1 validation built-in - automatically checks accessibility (alt text, color contrast, keyboard nav). Blocks completion until fixed.

• Load ANY Django project - not just VebGen-created ones. Migrate from Cursor mid-project.

• Improved code patching - 92% success rate (up from 70%) using fuzzy matching fallback.

• Dual agents - TARS plans, CASE codes, then they fix bugs together. 70% of errors auto-fixed.

• Production security - whitelist + blocklist + sandbox, no Docker needed.

THE NUMBERS:

900KB code | 356 tests (99.7% pass) | 1,000KB docs | $0 spent | MIT license

WHY DJANGO-ONLY?

Every tool tries to do everything. I wanted ONE framework done RIGHT - with real Django understanding, not generic autocomplete.

HONEST QUESTIONS:

1. Is AST + local parsing actually better than dumping code to LLMs? 2. Does built-in accessibility validation matter to you? 3. Should I go wider (FastAPI, Flask) or deeper (more Django features)?

The dual-agent approach adds latency but produces better code. Worth it?

GitHub: https://github.com/vebgenofficial/vebgen

Happy to answer technical questions!