Vector Database

A database optimized for storing and searching high-dimensional vectors — the embeddings used in semantic search and RAG.

A vector database stores embeddings — numerical representations of text, images, or other content — and lets you search by similarity rather than exact match. The core operation: given a query embedding, return the K nearest stored vectors using approximate nearest-neighbor (ANN) algorithms.

For agents, vector databases are the storage layer behind RAG and agent memory. When the agent needs context, you embed the query, search the vector DB, and inject the top results.

Popular 2026 options:

• Pinecone, Weaviate, Qdrant, Milvus — purpose-built vector databases with rich filtering
• pgvector (PostgreSQL extension) — pragmatic choice for teams already on Postgres
• Cloudflare Vectorize, Turbopuffer — serverless, pay-per-query economics
• Chroma, LanceDB — local/embedded options for development and small deployments

The vector DB choice rarely dominates agent quality — embedding model choice, chunking strategy, and retrieval pipeline design matter more. But the DB choice does dominate ops: scaling characteristics, query latency, filtering capabilities, and operational overhead differ widely. For most B2B teams, pgvector is the right starting point until volume or latency drives a specialized choice.