Code Review Graph
Local code intelligence for AI coding assistants. Builds an MIT-licensed repository graph and supplies focused review context through MCP or a CLI.
Code Review Graph builds a persistent map of your repository so AI coding assistants can review a change without repeatedly reading the whole codebase. It's an open source, MIT-licensed tool for developers and teams using Claude Code, Cursor, Codex, Copilot or other MCP clients. The graph and core analysis run on your machine.
It maps functions, classes, imports and their relationships, then traces the callers, dependencies and tests a change could affect. That gives an assistant a smaller set of relevant files to inspect. Tree-sitter parsing and a local SQLite database keep the map available between reviews; incremental updates process changed files rather than rebuilding everything.
Beyond affected files, the analysis identifies risky functions, execution flows and gaps in test coverage. It can also find unexpected coupling and code that connects otherwise separate parts of a project. An interactive graph lets you explore those relationships visually, while a GitHub Action runs the analysis on your CI runner and posts risk findings on pull requests without sending source code to an external analysis service.
Language support includes Python, TypeScript, Rust, Swift and Java, alongside Jupyter and Databricks notebooks. Framework-specific analysis follows Laravel relationships and Spring application flows.
Optional semantic search can use local sentence-transformers embeddings or cloud providers such as Google Gemini and Voyage AI. OpenAI-compatible endpoints include vLLM and LocalAI. Cloud embedding choices send data to the selected provider, and connected AI assistants handle supplied review context under their own data policies.
