Skip to main content

Glossary · ai

Vector Database

Definition

A vector database stores text/images/audio as high-dimensional vectors (1024-3072 dim) and runs similarity search. It's the backbone of modern AI applications like RAG, semantic search, recommendations, and AI agents. Pinecone, Qdrant, Weaviate, pgvector are leading choices.

Published: 2026-05-05Updated: 2026-05-05

Detailed explanation

Classic DBs do exact match (WHERE name = 'X'); vector DBs do semantic similarity. A query for 'dog' returns high similarity for 'cat' because their embeddings are close. This is the foundation of semantic search.

Indexing algorithms: HNSW (Hierarchical Navigable Small World — most common), IVF (Inverted File), FAISS. Trade-offs: speed vs accuracy vs memory.

2026 options: Pinecone (managed, most mature, $70+/month), Qdrant (Rust-fast, open source), Weaviate (built-in modules), pgvector (Postgres extension — leverages existing DB), Chroma (POC).

Use cases

RAG (document retrieval for LLM context)

Semantic search (meaning, not keywords)

Recommendation engines (user + product similarity)

Image / video / audio search

Anomaly detection (outlier vectors)

Pros

  • +Strong semantic search
  • +Multi-modal (text + image + audio)
  • +Horizontally scalable
  • +Core of modern AI infrastructure

Cons

  • Embedding cost (OpenAI $0.13/1M tokens)
  • Storage: 1M chunks × 3072 dim = ~12 GB
  • Re-embedding needed (new model)
  • Complexity (chunk + embed + retrieve + re-rank)

Related terms

EmbeddingRAGHNSWSimilarity Search

Related services

Planning a project around Vector Database?

In a 30-minute discovery call we share a written architecture + cost + team recommendation tailored to your project.

Start a discovery call