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! |