HNSW Algorithm

What Is HNSW Algorithm?

Pinecone, Qdrant, Weaviate, pgvector (with hnsw extension), and Milvus all default to HNSW. The index is a multi-layer graph: top layers are sparse for coarse search, bottom layers are dense for refinement. Trade-off: faster + more memory than IVFFlat, slower index build than flat. For most RAG workloads above 100k vectors, HNSW is the right default.