← Glossary/Architecture

Defined term

Vector store

A database optimized for similarity search over embeddings.

A vector store indexes embeddings using algorithms like HNSW or IVF for sub-linear nearest-neighbor search. Choices include managed services (Pinecone, Weaviate Cloud), open-source (Qdrant, Weaviate, Milvus), and Postgres extensions (pgvector). The right choice depends on scale, latency target, hybrid search needs, and operational constraints.

When it matters

When you have 10,000+ documents to retrieve over and your query volume justifies dedicated infrastructure. Below that, pgvector in your existing Postgres often beats specialized vector DBs.

Real example

A 4M-vector store on Pinecone for a legal-research workflow: 99.9% uptime, P99 query latency 45ms, 12 metadata filters per query (jurisdiction, date, topic, doc-type). Reranking on top-100 results before final answer generation.

KPIs to watch

Query latency P95 (<100ms target), retrieval recall@10 (>0.85), index refresh cycle time (<4h for daily ingest).

Related terms

See it in action

We use this every week

Book a 30-min call and we'll walk you through how Vector store shows up in a real engagement we're running.

Book a 30-min call