Unverified50% confidenceReleaseExact time
Since 0.8.0, pgvector provides an iterative index scan, enabled via SET hnsw.iterative_scan = relaxed_order or strict_order, which continues expanding the scan when candidates are exhausted until enough rows for the LIMIT are collected or the hnsw.max_scan_tuples limit is reached.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/17/2026
First Seen
Valid until: 1/15/2027
Sources
Related Entities
Related Claims
Unverifiedpgvector applies WHERE filtering only after the index scan; its README notes that if the filter condition matches 10% of rows, a top-10 query returns on average only 4 rows when the default hnsw.ef_search is 40.68% similarUnverifiedPg_vectorsearch does not follow the classic IVF or HNSW route, but instead designs a new index structure, focusing on optimizing query latency, index build speed and memory efficiency.63% similarUnverifiedAccording to benchmark data provided by the project, Pg_vectorsearch surpasses pgvector's IVFFlat and HNSW in query throughput.62% similarUnverifiedWith pgvector you can use ordinary WHERE clauses to restrict by tenant, permission, language, or date and then sort the remaining rows by similarity, whereas many standalone vector databases struggle with this pattern or require additional metadata filtering features.61% similarUnverifiedpgvector can skip the index within a transaction via SET LOCAL enable_indexscan = off to get exact answers, and Qdrant can achieve exact retrieval via "exact": true.61% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1050509API
curl https://kongchang.com/api/v1/knowledge/claims/1050509MCP
get_claim(id=1050509)