← AI Terminology

HNSW - Hierarchical Navigable Small World

HNSW is a graph-based ANN index that links vectors in a hierarchical small-world graph for fast approximate nearest-neighbour search with strong recall.

It is the default index type in many vector databases.
Why It Matters in AI
HNSW offers excellent speed/recall and incremental inserts, making it the practical default for RAG corpora. Knowing M, efConstruction, and efSearch parameters is everyday vector-DB tuning.
Key Points
Aspect Description
Cons Memory-heavy vs some compressed IVF-PQ setups
Pros High recall, good latency, dynamic inserts
Params M (degree), efConstruction, efSearch
Related ANN, FAISS HNSW indexes
Used in Milvus, Qdrant, Weaviate, pgvector HNSW, hnswlib
Structure Multi-layer proximity graph; greedy search from top layer
Simple Analogy
A multi-level highway system: zoom on long-range links first, then exit onto local roads to the exact address.
Common Usage Examples
  • efSearch=64 for higher recall
  • Qdrant/Milvus HNSW collections
  • hnswlib Python bindings
  • Memory plan: graph edges dominate RAM
Summary
In short: HNSW is the hierarchical graph index powering most production vector search — fast, high-recall approximate nearest neighbours.