Code-Graph-RAG

Code-Graph-RAG parses a multi-language codebase with Tree-sitter, sharpened by compiler-grade frontends and runtime traces where available, builds a knowledge graph of its structure in an embedded graph database that needs no server, and lets you query, edit, and optimise that code in plain English. It works across a monorepo of mixed languages under one unified graph schema.

Latest News 🔥

• Type Handling: Improved handling of type context managers, walrus operators, and match binding targets in both Python and TypeScript.
• Rename Refinement: Prevented Go renames from unexporting names used by other packages.
• Profile Naming: Corrected the naming of pprof profiles to avoid "malformed" errors when no samples are present.
• Output Budget: Increased the Anthropic output budget to improve the default thinking experience.
• Module Resolution: Resolved issues with module resolution in various scenarios, including ES module default imports and bare name bindings.
• Dead Code Elimination: Removed dead code roots associated with IPython rich-display methods.
• Incremental Refresh: Improved incremental refresh to ensure callers are updated when dispatch targets change.

See NEWS.md for the full history.

What It Does

Point Code-Graph-RAG at a repository and it reads every source file, extracts functions, classes, methods, modules, and the relationships between them, and stores the result as an interconnected graph. Once the graph exists you can:

• Ask questions about the codebase in natural language and get answers grounded in the real structure.
• Retrieve the actual source of any function, class, or method by name or by intent.
• Edit code through the agent with AST-based surgical patching and a diff preview before anything changes.
• Optimise code against language best practices or your own coding standards.
• Find dead code by walking call and reference edges from entry points.
• Search and rewrite structurally by AST pattern with ast-grep.
• Overlay runtime behaviour: trace a test run (or pull production eBPF profiles) with cgr trace and merge the calls that actually happened into the graph, exposing dispatch that static analysis cannot see.

How It Works

The system has two components:

1. Multi-language parser. A Tree-sitter based parser reads the codebase and ingests functions, classes, methods, modules, and their relationships into the graph under a single language-agnostic schema. Where a toolchain is available, compiler-grade frontends layer exact facts on top (libclang for C/C++, go/types for Go, and opt-in Roslyn, javac and Jedi for C#, Java and Python), and dynamic tracing merges calls observed at runtime. Tree-sitter stays the backbone: a trace only sees code that ran, and a compiler frontend only covers what its toolchain can build.
2. RAG system (codebase_rag/). An interactive CLI that turns natural language into Cypher queries, retrieves matching code, and drives AI-powered editing and optimisation.

Source Code -> Tree-sitter Parser -> AST Analysis -> Knowledge Graph
                                                             |
User Query -> AI Model (Cypher Gen) -> Cypher Query -> Graph Results -> Response

The graph is stored in an embedded SQLite file by default. Memgraph and Neo4j are opt-in backends that run the same Cypher; Graph Backends compares them.

See the Architecture Overview and Graph Schema for the full picture.

Supported Languages

Python, TypeScript, TSX, JavaScript, Rust, Go, Java, C, C++, C#, PHP, Lua, Dart, and Julia are fully supported. Scala is in development, and Ruby, Kotlin, Swift, Elixir, Haskell, Solidity, Bash, and Nix have structural support (modules, functions, classes where the language has them, and imports) through the pluggable ast-grep tier. See the Language Support matrix for per-language capabilities.

Installation

cgr is published to PyPI. Install it system-wide with the treesitter-full (tree-sitter grammars), ast-grep (Ruby, Kotlin, Swift, Elixir, Haskell, Solidity, Bash, Nix) and semantic (vector search) extras:

# with uv (recommended)
uv tool install "code-graph-rag[treesitter-full,ast-grep,semantic]"

# or with pipx
pipx install "code-graph-rag[treesitter-full,ast-grep,semantic]"

Which version am I getting?

Three version lines exist and they intentionally differ:

where	what it tracks
git tags	every version, one per merge
GitHub Releases (binaries, signatures)	every 50th version, plus any security fix, breaking change or merge labelled release now
PyPI	every 50th version, plus any security fix, breaking change or merge labelled release now

So the newest tag on main usually runs ahead of the newest release, often by tens of patch versions; they coincide only just after a release. Nothing is stuck, the cadences differ by design. A security fix, a breaking change or a merge a maintainer labels release now does NOT wait for the cadence: each ships a release and a PyPI upload immediately.

uv tool install and pipx install give you the newest PyPI version, which is the newest RELEASE, not the newest tag. Interim tags exist so every merge is addressable; binaries and PyPI uploads follow the cadence above.

To run code newer than the latest release, install from git:

uv tool install "code-graph-rag[treesitter-full,ast-grep,semantic] @ git+https://github.com/vitali87/code-graph-rag@main"

To upgrade an existing install, run uv tool upgrade code-graph-rag (with pipx, pipx upgrade code-graph-rag, or pipx reinstall code-graph-rag for a git install). Upgrade covers the other install methods.

This is the beginning of the README. Read the rest at the original. Read the full README ↗