LxRAG

About

LxRAG functions as an MCP-native memory and code intelligence layer, enhancing LLM agents with smarter, faster AI-assisted development capabilities. It transforms code repositories into queryable graphs, enabling agents to answer complex architecture, impact, and planning questions without repeatedly parsing the entire codebase. By uniquely combining graph structure, session persistence for agent memory, and hybrid retrieval methods (graph + vector + BM25), LxRAG eliminates context loss, improves architectural understanding, and facilitates temporal reasoning for code, all while offering 38 deterministic tools for code intelligence.

Key Features

• Impact-scoped Test Selection: Identifies and runs only the relevant tests affected by code changes, significantly accelerating development cycles.
• Hybrid Retrieval: Combines graph traversal, vector similarity, and BM25 lexical search with Reciprocal Rank Fusion for highly accurate natural-language code queries.
• MCP-native Runtime: Seamlessly integrates with tools like VS Code Copilot and Claude through a standardized JSON-RPC 2.0 interface.
• Persistent Agent Memory: Stores agent decisions, episodes, and code states, allowing temporal queries and surviving session restarts.
• 1 GitHub stars
• Code Graph Intelligence: Provides cross-file dependency answers and architectural understanding by representing code as a queryable graph.

Use Cases

• Enabling LLM agents to answer complex questions about codebase architecture and dependencies in natural language.
• Maintaining persistent context and memory for AI agents across multiple development sessions and restarts.
• Optimizing software testing by precisely identifying and running only the tests impacted by recent code modifications.