I built this because AI coding assistants lose context across large codebases. They understand individual files well, but structural questions ("what calls this function?", "show the inheritance chain for BaseHandler") require reading many files sequentially.

Code-Graph-RAG parses your codebase into an AST with Tree-sitter, builds a knowledge graph in Memgraph, and exposes it to AI assistants through MCP. A graph traversal gives precise answers where vector similarity search gives approximations.

It supports 11 languages with a unified schema, so polyglot monorepos work out of the box. The Cypher generation layer can use OpenAI, Gemini, or fully local Ollama.

Docs: https://docs.code-graph-rag.com