hnsw
technique judged 100% likely a named technique or architecture
See how it has spread across media tiers
Mentions over time
Table view
| Week | Articles | Sources |
|---|---|---|
| 2024-W02 | 1 | 1 |
| 2024-W26 | 1 | 1 |
| 2024-W40 | 1 | 1 |
| 2024-W48 | 1 | 1 |
| 2025-W23 | 1 | 1 |
| 2025-W49 | 1 | 1 |
| 2026-W11 | 5 | 1 |
| 2026-W13 | 1 | 1 |
| 2026-W24 | 1 | 1 |
| 2026-W38 | 1 | 1 |
| 2026-W40 | 2 | 2 |
| 2026-W41 | 1 | 1 |
In their words
collection_name , vectors_config = VectorParams ( size = vector_size , distance = distance , hnsw_config = HnswConfigDiff ( m = 16 , # Connections per layer ef_construct = 100 # Build-time accuracy ) ) ) def upsert_chunks ( self , chunks : List [ Chunk ], embeddings : List [ List [ float ]]): points = [ PointStruct ( id = str ( uuid .
For larger ones, approximate nearest neighbor indexes such as HNSW or IVF trade a little recall for large speedups.
For agents that rely on vector search, fast approximate nearest neighbor indexing such as HNSW enables low-latency retrieval even as memory grows .
This allows us to keep only the small HNSW (Hierarchical Navigable Small World) graph and compressed vectors in memory, while reading a small number of full-precision vectors from disk for rescoring.
The dominant index type is HNSW (Hierarchical Navigable Small World), a graph-based index used by Qdrant, Weaviate, and Pinecone.
Vector search has evolved Production vector search is built from independent primitives that interact, not a single retrieval call: - Quantization for memory compression - HNSW parameter tuning for recall and latency tradeoffs - Payload filtering with indexed fields - Fusion strategies for hybrid retrieval (dense + sparse) - Sharding and replication for scale Each one of these is a lever.
- Vector database configuration — index type, HNSW parameters, and filtering strategies.
Most vector databases use HNSW (Hierarchical Navigable Small World) indexing, a graph-based ANN algorithm that provides excellent speed-quality tradeoffs.