ComponentsBuild with this
rag-graphmainstream
GraphRAG
Knowledge graph-based retrieval for multi-hop reasoning.
Overview
Builds a knowledge graph from your corpus and uses graph traversal for retrieval. Excels at queries that require connecting entities across multiple documents — relationships, causal chains, and cross-document reasoning.
Architecture
Interactive walkthrough01
Indexing
Document Loadercorpus intake
Entity ExtractionLLM-powered NER
Relationship Miningsubject → predicate → object
Knowledge GraphNeo4j
02
Retrieval
Query → CypherLLM translation
Multi-hop TraversalA → B → C paths
Subgraph Contextentities + edges
03
Generation
Graph Verbalizerpaths → prose
LLM Callreasoning over relations
Answer + Pathstraceable hops
01
Indexing
Document Loadercorpus intake
Entity ExtractionLLM-powered NER
Relationship Miningsubject → predicate → object
Knowledge GraphNeo4j
02
Retrieval
Query → CypherLLM translation
Multi-hop TraversalA → B → C paths
Subgraph Contextentities + edges
03
Generation
Graph Verbalizerpaths → prose
LLM Callreasoning over relations
Answer + Pathstraceable hops
Summarized pipeline view. For the full interactive, scroll-driven walkthrough with clickable stages → Pipeline detail
When to use
Use when
- multi-hop reasoning required
- queries involve entity relationships
- corpus has structured knowledge (people, orgs, events)
- high-value queries justify higher indexing cost
Avoid when
- corpus updates in real-time (graph rebuild is expensive)
- simple factual Q&A
- team has no graph infrastructure experience
- latency sensitivity < 500ms p50
Compatible vector databases
Neo4jQdrantPinecone
Compatible frameworks
langchainllamaindexraw python
#graph#knowledge-graph#multi-hop#entity-linking#neo4j
Ready to build with GraphRAG?
Walk through the wizard to generate a complete, parameterized pipeline.