GraphRAG: Retrieval with Structure

Baseline RAG retrieves chunks — blind to who relates to whom across the corpus. GraphRAG adds an indexing pass: extract entities and relations from the documents into a graph, then at query time retrieve subgraphs, paths, and community summaries instead of (or alongside) raw text.

The corpus becomes queryable as structure: 'how do these two companies connect' becomes path-finding instead of hoping one chunk happens to name both. The extraction cost is real and upfront — you're paying indexing time to buy a new question shape.