# codesight

A universal AI context generator that maps a codebase into a compact, AST-based summary, saving thousands of tokens per conversation in Claude Code, Cursor, Copilot, Codex and similar tools.

codesight analyzes a project once and produces a compact context file that an AI coding assistant can read instead of exploring the codebase file by file, cutting the tokens spent figuring out a project's structure. It has zero runtime dependencies, detects more than 30 frameworks and 14 ORMs, and parses TypeScript and JavaScript projects with full AST precision, falling back to regex-based detection for Python, Go, Ruby, Elixir, Java, Kotlin, Rust, PHP, Dart, Swift, C# and BrightScript.

A wiki mode compiles the same AST data into a persistent set of short, topic-specific Markdown articles, such as `auth.md` or `database.md`, committed to the repository so each new agent session can read one small article instead of reloading a full context map; a watch mode keeps the wiki current as the code changes, and a git-hook mode regenerates it on every commit. It exposes 14 MCP tools and installs with a single `npx codesight` call.

Website: https://github.com/Houseofmvps/codesight