vitali87/code-graph-rag

Code‑Graph‑RAG – AI‑powered code‑base knowledge graph

What it is – An open‑source tool that parses a multi‑language repository with Tree‑sitter, stores the extracted functions, classes, modules and their relationships in a Memgraph knowledge graph, and lets you interact with that graph through natural‑language queries. It also supports AI‑driven code editing, optimisation, dead‑code detection and runtime‑trace overlay.

How it works

1. Parsing – Tree‑sitter walks every source file (Python, TS/JS, Rust, Go, Java, C/C++, C#, PHP, Lua, Dart, etc.) and builds an AST.
2. Graph ingestion – The AST is transformed into a language‑agnostic schema and loaded into Memgraph.
3. RAG loop – A CLI (cgr) turns a user’s English question into a Cypher query via an LLM, runs it against the graph, and returns a grounded answer. The same loop can generate AST‑based patches for edits.

Key capabilities

• Natural‑language Q&A over the whole codebase, with results tied to real symbols.
• Source retrieval by name or intent.
• AI‑driven editing: surgical AST patches previewed as diffs before applying.
• Code optimisation against best‑practice rules or custom standards.
• Dead‑code detection by traversing call/reference edges.
• Structural search & rewrite using ast‑grep patterns.
• Dynamic tracing: merge eBPF or test‑run call graphs into the static graph to expose runtime dispatch.
• MCP server compatibility so tools like Claude Code can query/edit the graph directly.

Installation & usage

• Install the cgr CLI from PyPI with the full Tree‑sitter and semantic extras (pipx or uv).
• Run the bundled Memgraph + Qdrant stack with cgr daemon up (Docker required).
• Index a repo: cgr start --repo‑path /path/to/repo --update‑graph.
• Query or edit via the interactive CLI or programmatically through the Python SDK.

Why it matters

• Provides a single, language‑agnostic graph for monorepos that mix dozens of languages, something most code‑search tools lack.
• Bridges static analysis and runtime information, giving a more complete picture of how code actually behaves.
• Enables AI agents to make precise, AST‑level changes rather than crude text edits, reducing the risk of breaking code.
• Open‑source core with optional enterprise‑grade managed or air‑gapped deployments, making it suitable for regulated environments.

Resources

• Detailed docs: installation, quick‑start, CLI reference, architecture, SDK, and advanced topics.
• Community: CI badge, code‑coverage, SonarCloud quality gate, OpenSSF scorecards.
• Enterprise services for managed cloud or on‑premise hosting.

Bottom line – Code‑Graph‑RAG is a full‑stack, AI‑enhanced platform for turning any codebase into a queryable knowledge graph, enabling natural‑language insight, safe automated refactoring, and deeper static‑dynamic analysis.