Hey HN, I got incredibly tired of RAG for code. Vectorizing source code into arbitrary chunks and hoping a cosine-similarity search finds the right structural context is a black box that just doesn't reliably work for agentic coding. It's anti-KISS.

So I wrote AstrMap in Go.

It rips through your repo in milliseconds and generates a highly compressed, AI-readable AST "Map" (like a giant table of contents with line numbers and functions). You feed the Map to your LLM (costs almost 0 tokens), and the LLM uses it as a deterministic radar to pinpoint exactly which files it needs to modify. It treats your architecture as pure, parseable text.

It's completely free, a single binary, local, and supports Go, JS/TS, Python, HTML/CSS. Give it a try and let me know if it speeds up your agentic coding workflows.

Link: https://github.com/hubby247/astrmap