The GraphRAG patterns described in this text are a set of retrieval strategies for advanced RAG systems that leverage graph structures for more effective retrieval. The most basic pattern, the Basic Retriever, uses vector similarity search on chunk embeddings to retrieve relevant chunks. Intermediate patterns like Parent-Child Retriever and Hypothetical Question Retriever build upon this by incorporating additional context or relationships within the data. Advanced patterns like Graph-Enhanced Vector Search and Global Community Summary Retriever use graph structures to provide more comprehensive context for answering questions. Each pattern has its own set of required pre-processing steps, graph patterns, and retrieval queries, making it essential to experiment with different patterns to find the most suitable one for a specific application. The journey to discovering ideal GraphRAG patterns is ongoing, filled with trial, error, and innovation.