zzszmyf/codefuse

CodeFuse is a tool that pre-indexes a codebase so that AI coding assistants can find and understand code faster and more accurately. The problem it addresses is that most AI coding tools currently navigate code by running text searches across files every time they need to find something. This is slow, imprecise, and produces results that include false matches from comments or unrelated code. Instead of searching files on demand, CodeFuse reads the entire codebase once and builds a graph that tracks every function, class, and method along with which ones call which others. That graph is then stored locally. When an AI agent needs to find a symbol, look at who calls a particular function, or see the structure of a file, it queries the pre-built index rather than scanning raw text. The result is faster lookups and more accurate results, particularly for cross-file relationships that are invisible to plain text search. The tool exposes the index in a few different ways. The command line accepts queries for symbol names with support for wildcards, so you can search for everything matching a pattern like "get*User". It can also generate a folder of plain text files that mirrors the index structure, so an AI agent can navigate code using ordinary file-listing and file-reading operations without any special integration. Optionally it can mount the index as an actual filesystem using FUSE, letting you browse symbols the same way you browse a directory tree. Indexing supports Go, TypeScript, JavaScript, Python, and Rust. The fast default mode uses the Go compiler's own parser for Go files and regular expressions for other languages. A slower but more thorough mode uses tree-sitter, an open-source code parsing library, for more accurate results across all languages. The index is stored in a project subfolder and supports incremental updates, so only changed files are re-parsed on subsequent runs.