Glossary · Foundations

Vector Database

A vector database stores embeddings and finds similar items by approximate nearest-neighbor search. Definition, top vendors, and when you actually need one.

By Kadin Nestler · May 28, 2026 · Updated May 28, 2026
On this page

Why vector databases exist

Traditional databases find rows by exact match. "Find the customer named Jane Doe." Vector databases find rows by meaning. "Find the customer notes most semantically similar to this complaint." The query is a vector. The match is approximate nearest neighbor (ANN) — algorithmically close, not literally identical. ANN at scale was the bottleneck that made retrieval-augmented LLM applications impractical until purpose-built vector engines arrived.

Leading vector databases in 2026

  • Pinecone — managed, fully cloud, popular for production RAG.
  • Weaviate — open-source plus managed, hybrid search (vector + keyword) built-in.
  • Qdrant — open-source, Rust-based, strong filtering.
  • Chroma — open-source, easy local dev, popular for prototypes.
  • pgvector — Postgres extension. If you already run Postgres, often the right answer.
  • Milvus / Zilliz — open-source plus managed, billion-scale workloads.

When you need a dedicated vector database

Under 100K vectors: in-memory FAISS or pgvector is plenty. Over 1M vectors with sub-100ms query latency requirements: dedicated managed vector DB. Multi-tenant SaaS with thousands of small indexes: Pinecone or Weaviate cloud handle isolation cleanly. If you are already in Postgres for the rest of the app, pgvector usually wins on operational simplicity until you hit scale issues.

What an SMB owner should know

  • Embedding model and vector DB must be paired — switching embedding models means re-indexing everything.
  • Cost is usually dimension count times vector count times queries per month. Plan for growth.
  • Hybrid search (vector + keyword) usually beats pure vector for recall on real workloads.
  • Vendor lock-in is real. Choose a database that supports export to standard formats.

What it means for your business

You probably do not buy a vector database directly. Your AI vendor does. But if they cannot tell you which one they use and why, they are likely guessing — and you will pay for that guess in retrieval quality six months in.

  • Embedding — An embedding is a numeric vector that represents the meaning of text, an image, or audio. Definition, top embedding models, and how they power search.
  • Retrieval-Augmented Generation (RAG) — RAG is the technique of fetching documents from a database and feeding them to an LLM before it answers. Definition, architecture, and SMB use cases.
  • AI Knowledge Base — An AI knowledge base is a structured corpus of documents an AI agent retrieves from to answer questions. Definition, architecture, and SMB setup tips.
  • AI Grounding — Grounding is the practice of tying AI outputs to verified source material. Definition, techniques, and why it is the primary defense against hallucination.
  • Large Language Model (LLM) — A Large Language Model is a transformer-based neural network trained on trillions of tokens to predict the next token. Definition, key models, and business use.
← Back to the full glossary