Graph Databases vs Vector Databases

A vector index (HNSW and friends) answers 'what else looks like this' — approximate nearest neighbors in embedding space. A graph index answers 'what is connected to this, and how' — adjacency lists with typed edges. One is geometry, the other is topology.

'Find documents like this paragraph' is a vector question. 'Which suppliers feed a part that goes into products sold in Europe' is a graph question — three joins, zero similarity. Neither index can fake the other's job.