neo4j-graphrag-skill

What problem does it solve?

GraphRAG retrieval in Neo4j can be difficult to wire correctly, especially when choosing the right retriever type, creating the required indexes, and composing safe Cypher-based graph traversal after vector search.

Core Features & Use Cases

• Retriever selection & configuration: Choose the right retriever (VectorRetriever, HybridRetriever, VectorCypherRetriever, HybridCypherRetriever, Text2CypherRetriever, ToolsRetriever, and external vector retriever variants) based on whether you have fulltext, need graph traversal, or want LLM-routed retrieval.
• Graph-augmented retrieval logic: Define a retrieval_query Cypher fragment with automatic node and score injection, plus parameterized query_params and pre-filtering via filters.
• Production-grade pipeline wiring: Assemble GraphRAG(retriever=..., llm=...) and run .search() with support for many LLM and embedder providers, plus retrieval quality debugging with return_context and response_fallback.
• Index prerequisites & compatibility: Guides vector + fulltext index setup for hybrid retrievers, supports neo4j-graphrag v1.7+ (neo4j-genai rename), and aligns with Neo4j driver/Neo4j version requirements.

Quick Start

Install neo4j-graphrag, create the required vector (and fulltext for hybrid) indexes, write a retrieval_query that returns node and score, then run GraphRAG.search to get graph-augmented answers.