Leveraging Tree-sitter, Better Code Review Graph parses codebases to build a detailed structural knowledge graph of functions, classes, and imports. This enables AI agents, such as Claude, Cursor, and Copilot, to perform token-efficient code reviews by supplying them with highly relevant and precise context. It addresses critical limitations found in its upstream fork, offering improved search capabilities, reliable call resolution, and configurable dual-mode embeddings for flexible performance, either locally via ONNX or through cloud providers via LiteLLM. The tool optimizes context size with paginated outputs, ensuring AI models receive only necessary information.
Key Features
01AND-logic multi-word search for precise code entity discovery
02Dual-mode embedding support with local ONNX (qwen3-embed) or cloud LiteLLM integration
031 GitHub stars
04Generates a structural graph of code from various programming languages using Tree-sitter
05Qualified name resolution for accurate identification of callers and callees
06Paginated output for queries to control context size for AI agents
