← AI Terminology
Vector Database
A vector database stores embedding vectors and supports efficient similarity search (ANN) so applications can retrieve nearest neighbours by semantic meaning.
It is core infrastructure for RAG, recommendation, and multimodal search.
It is core infrastructure for RAG, recommendation, and multimodal search.
Why It Matters in AI
LLMs need external memory; embeddings make meaning searchable. Vector DBs (and vector indexes in Postgres/Elastic) operationalise semantic retrieval at scale with filters, hybrid search, and CRUD.
Key Points
| Aspect | Description |
|---|---|
| Ops | Insert embeddings, ANN query, metadata filters |
| Hybrid | Often combined with BM25 keyword search |
| Engines | Pinecone, Weaviate, Milvus, Qdrant, Chroma, pgvector |
| Related | Embedding models, RAG, semantic search |
| At scale | Sharding, replication, quantised vectors |
| Index types | HNSW, IVF, DiskANN variants |
Simple Analogy
A library organised by topic similarity rather than only alphabetical title — books about related ideas sit near each other in a high-dimensional map.
Common Usage Examples
collection.query(query_embeddings=..., n_results=5)- pgvector
ORDER BY embedding <=> $1 LIMIT k - Pinecone/Weaviate production indexes
- RAG chunk store for enterprise docs
Summary
In short: A vector database indexes embeddings for fast similarity search — the storage engine behind most RAG systems.