Back to AI Glossary
Retrieval & Data

Vector Database

A database optimized for storing, indexing, and querying high-dimensional vector embeddings, enabling fast similarity search at scale.

Why it matters

Vector databases are the storage backbone of every RAG pipeline and semantic search system. Choosing the right one affects retrieval quality, latency, and infrastructure cost.

Why vector databases exist

Traditional databases index by exact values (IDs, keywords). AI applications need to find items by semantic similarity — "show me documents about X" rather than "show me documents containing the word X." Vector databases use approximate nearest neighbor (ANN) algorithms to make this fast even at millions of vectors.

Architecture choices

  • Dedicated vector DBs — Pinecone, Weaviate, Qdrant. Purpose-built, feature-rich, but another service to manage.
  • Extensions to existing DBs — pgvector (Postgres), Atlas Vector Search (MongoDB). Keep everything in one database, simpler ops.
  • In-memory libraries — FAISS, Annoy. Fast for small-to-medium datasets, no persistence.

What to look for

Key differentiators: filtering support (combine vector search with metadata filters), update speed, managed vs. self-hosted, cost per million vectors, and hybrid search (combining vector + keyword search).

Related Terms

Embeddings— Dense numerical representations of text (or other data) in a high-dimensional vector space, where similar meanings are placed closer together.Retrieval-Augmented Generation(RAG)— A technique that grounds a language model's output in external data by retrieving relevant documents before generating a response.

On the AI Radar

pgvector— An open-source PostgreSQL extension that adds vector similarity search with HNSW and IVFFlat indexing directly inside Postgres, allowing teams to store embeddings alongside relational data without a separate vector database.Pinecone— A fully managed, cloud-native vector database designed for similarity search at scale. Provides serverless infrastructure for storing and querying high-dimensional vector embeddings with sub-25ms latency, integrated inference for embedding and reranking, and dedicated read nodes for predictable performance.
Back to AI Glossary