Vector Database

Definition

A vector database is a database purpose-built to store, index, and search high-dimensional vector embeddings, retrieving items by semantic similarity rather than exact keyword match.

Overview

Machine learning models can convert text, images, audio, or other content into embeddings — dense numeric vectors that place semantically similar items close together in a high-dimensional space. A traditional database can store these vectors as columns, but finding the 'nearest neighbors' to a query vector across millions of rows using brute-force comparison is far too slow for real-time applications. Vector databases solve this with specialized indexing structures, most commonly variants of Approximate Nearest Neighbor (ANN) algorithms like HNSW (Hierarchical Navigable Small World graphs) or IVF (Inverted File indexes), which trade a small amount of accuracy for dramatic speed gains. Beyond fast similarity search, production vector databases typically support metadata filtering (combining a semantic search with structured filters like date or category), hybrid search that blends vector similarity with traditional keyword search, and horizontal scaling to handle billions of vectors across distributed nodes. Vector databases are the retrieval backbone of RAG (retrieval-augmented generation) systems, where relevant document chunks are embedded, stored, and retrieved to ground an LLM's responses in real, up-to-date information rather than relying solely on what the model memorized during training. They are also used for recommendation systems, image and audio similarity search, anomaly detection, and deduplication. Purpose-built options include Pinecone, Weaviate, Qdrant, Milvus, and ChromaDB, while several traditional databases (PostgreSQL via the pgvector extension, Elasticsearch, Redis) have added vector search capabilities rather than requiring a dedicated system. Building RAG applications and choosing the right vector store is covered in depth in Retrieval-Augmented Generation, and the tradeoffs between options are explored further in Vector Databases Explained.

Key Features

• Stores and indexes high-dimensional vector embeddings for similarity search
• Uses Approximate Nearest Neighbor (ANN) algorithms like HNSW or IVF for fast retrieval at scale
• Supports metadata filtering to combine semantic search with structured constraints
• Often supports hybrid search blending vector similarity with keyword search
• Scales horizontally to handle billions of vectors across distributed nodes
• Core retrieval layer for RAG pipelines grounding LLM responses in real data
• Available as dedicated systems (Pinecone, Weaviate, Qdrant, Milvus) or extensions to existing databases (pgvector)

Use Cases

Frequently Asked Questions

Related in Database