XNews

codesight: Static Codebase Docs for AI Assistants, Zero API Cost

codesight: Static Codebase Docs for AI Assistants, Zero API Cost

codesight pre-compiles a codebase into structured documentation—routes, data flows, and module relationships—using pure static analysis without calling any large language model. This produces a table of contents that an AI assistant can read in about 200 tokens to understand the project, then navigate to specific pages for details like authentication logic. The generation takes 200 milliseconds and incurs zero API cost. Inspired by Karpathy's LLM wiki concept, codesight derives documentation directly from code structure rather than relying on model summarization. It also generates configuration files like CLAUDE.md and AGENTS.md, and can show which parts of the codebase are affected before making changes. Compatible with Cursor and Codex, it benefits larger repositories most.

Source posts