graphsub.com / wiki / graph-rag
GraphRAG and Source-Linked Retrieval
GraphRAG combines graph structure with retrieval-augmented generation so AI systems can answer with entities, relationships, provenance, and source context.
Definition
GraphRAG uses graph relationships to improve retrieval and generation. Instead of retrieving isolated chunks, the system can traverse entities, claims, events, owners, and sources.
GraphSub role
GraphSub supports GraphRAG-style workflows by storing memory, source links, tool traces, and output lineage as graph state.
Enterprise value
Source-linked retrieval matters when teams need answers that can be inspected, replayed, and connected to operational systems.
Code example
Source-linked GraphRAGtypescript
const answer = await graphsub.retrieve({
query: "Which vendor changes affect the risk replay plan?",
traverse: ["Company", "Claim", "Source", "Output"],
requireSources: true,
maxHops: 4,
})
for (const citation of answer.sources) {
console.log(citation.url, citation.claim_id)
}Retrieve graph context with entities, claims, and source links.
FAQ
- How is GraphRAG different from vector search?
- Vector search finds semantically similar content; GraphRAG adds explicit relationships and provenance.
- Does GraphSub include vector search?
- The product surface includes vector and graph-oriented retrieval lanes for agent workflows.