A vector database is a type of database purpose-built for storing and querying vector embeddings, the numerical representations that AI models use to encode meaning. Unlike a traditional relational database that matches records by exact field values, a vector database matches by semantic similarity: given a query, it finds the stored items that are conceptually closest to the query's meaning, even if they share no words in common.
Vector databases are a core infrastructure component of retrieval-augmented generation (RAG) systems. When a company's documents, policies, product catalog, or knowledge base are processed into embeddings and stored in a vector database, an AI assistant can retrieve the most relevant passages at query time and use them to generate accurate, grounded responses. Without a vector database, a language model can only answer questions based on what it learned during training.
Common vector databases include Pinecone, Weaviate, Qdrant, Chroma, and pgvector, which adds vector search capabilities directly to PostgreSQL. The right choice depends on data scale, latency requirements, whether the organization prefers a managed cloud service or self-hosted infrastructure, and whether vector search needs to coexist with traditional relational queries.
Performance in a vector database depends on the quality of the embeddings stored in it, which in turn depends on the embedding model used to generate them and how the source documents were prepared and chunked before processing. A well-architected vector database layer is one of the primary determinants of AI assistant quality in enterprise deployments.