At The Muses, we're researching adaptive learning and human-computer interaction using LLMs and non-invasive BCI technology. Our engineers work across large, interconnected codebases daily, and we kept hitting the same friction: AI coding assistants are powerful, but they lose context the moment a problem spans more than a few files. So we built GraphRepo for them. The idea is simple. Instead of feeding an assistant files one at a time, give it a queryable knowledge graph of your entire codebase. Parse a repo and get a full graph of every function, class, import, and call relationship. The assistant asks questions about structure instead of reading files sequentially. Everything runs locally on your machine. No Docker, no database servers, no API keys, no network calls. Your code never leaves your device. For teams working on proprietary or sensitive codebases, that was non-negotiable for us and we expect it is for you too. We added community detection (Louvain) and semantic search (Transformers.js, fully local) because we found those were the queries that actually changed how the assistant reasons about code. It supports TypeScript, JavaScript, Python, C, C++, C#, and Swift with full AST-level parsing via Tree-sitter. Everything else gets indexed at the file level. Install from the VS Code Marketplace, parse your workspace, and your assistant picks up the MCP tools automatically. Zero config. We'd love to hear what languages and features to prioritize next. Open source, built in TypeScript, and available now. Feel free to try it out, or contribute if you'd like to add any features or language support. GitHub: https://github.com/the-muses-ltd... VS Code Marketplace: https://marketplace.visualstudio...

GraphRepo

A knowledge graph for your codebase, queryable by AI via MCP

A knowledge graph for your codebase, queryable by AI via MCP