What problem does it solve? Building retrieval-augmented generation over a Neo4j knowledge graph requires choosing the right retriever, writing retrieval_query Cypher fragments, wiring embedders and LLMs, and managing indexes — this Skill provides the correct patterns and API usage for the neo4j-graphrag Python package so you avoid common misconfigurations. ## Core Features & Use Cases - Retriever Selection Guidance: Decision logic for VectorRetriever, HybridRetriever, VectorCypherRetriever, HybridCypherRetriever, Text2CypherRetriever, ToolsRetriever, and external vector DB retrievers (Weaviate, Pinecone, Qdrant). - Pipeline Wiring: Complete patterns for GraphRAG(retriever, llm) with prompt templates, message history, filters, query_params, return_context, and token usage tracking across OpenAI, Anthropic, VertexAI, Bedrock, Cohere, Mistral, and Ollama. - Index & Error Reference: Vector/fulltext index creation, schema inspection, and a table of common errors (missing score in RETURN, index not ONLINE, dimension mismatch) with fixes. - Use Case: You have Chunk nodes with embeddings in Neo4j and want a question-answering pipeline that traverses from matched chunks to related articles and organizations — use HybridCypherRetriever with a retrieval_query and GraphRAG.search(). ## Quick Start Ask the assistant to build a GraphRAG pipeline over your Neo4j database using HybridCypherRetriever with an OpenAI embedder and a retrieval_query that traverses from matched chunks to related entities.
