GraphRAG

GraphRAG is an architectural pattern and specific implementation (pioneered by Microsoft) that enhances standard RAG systems by incorporating knowledge graph structure into the retrieval and reasoning process. Rather than relying solely on vector similarity to find relevant documents, GraphRAG uses graph-based relationships between entities to guide retrieval and provide structured context.

The core GraphRAG approach involves: extracting entities and relationships from documents to build a knowledge graph, using community detection to identify topic clusters, generating summaries at different levels of the graph hierarchy, and leveraging this structure during query time to provide both local (specific entity) and global (high-level summary) context to the LLM.

GraphRAG is particularly effective for queries that require understanding relationships between entities, synthesizing information across many documents, or providing answers that span multiple topics. Microsoft's research showed that GraphRAG significantly outperforms standard RAG on questions requiring comprehensive understanding of a document corpus, such as "What are the main themes discussed across all documents?" The pattern has become influential in enterprise AI deployments where relationship understanding and explainability are critical.