LightRAG, an advanced knowledge retrieval system, combines two parallel retrieval paths to provide comprehensive and accurate answers. The hybrid approach blends graph traversal with semantic vector similarity, allowing for explainable and contextually grounded responses. It uses dual-level keyword extraction to understand both the big picture and important details in a user's query, prioritizing well-connected and central entities first. LightRAG retrieves structured knowledge from a graph and surrounding documents using dual-level keyword semantic search and graph traversal over entities and relationships. In parallel, it runs a semantic vector search with conversation history augmentation, retrieving semantically similar chunks from a pre-indexed vector store. The system combines the strengths of both retrieval methods, leveraging graph structure to prioritize information and providing more comprehensive responses to user queries.