Decoder. plain-English AI glossary

GraphRAG

▲ Rising

Also called Graph-based RAG

RAG that first builds a graph of entities and relationships, so it can answer questions plain chunk-search can’t.

Think of it like

Instead of skimming loose pages, drawing the org chart first so you can trace who reports to whom.

Example

Asked "how are these three suppliers connected?", GraphRAG walks the relationship graph rather than hoping one chunk mentions all three.

How it actually works

Standard RAG retrieves independent chunks, which struggles with questions that span many documents or need a global view. GraphRAG extracts entities and relations into a knowledge graph, often clusters it into summarized communities, and retrieves over that structure. It shines on "connect the dots" and whole-corpus questions, but building and maintaining the graph is real extra cost and complexity.

For product teams

Worth it when users ask synthesis questions across a big corpus; overkill for simple lookup FAQs.

For engineers

Entity/relation extraction into a graph plus community summaries, retrieved via graph traversal rather than flat top-k.

RAG — A structured extension of ordinary RAG.
Retrieval — Still relies on retrieval over the graph.

Read anything AI without the jargon

Look up any term in plain English, or save terms as you read with the free Chrome extension.

Open Decoder Add to Chrome