Vector Database
A database designed to store vector embeddings and retrieve the items most similar to a query vector at scale.
Vector databases are the storage and retrieval layer behind most RAG assistants and semantic search tools, so the choice affects answer quality, latency, cost and how easily access controls and data updates are enforced.
A legal tech firm stores 2 million clause embeddings in PostgreSQL with the pgvector extension and an HNSW index, filters by jurisdiction and document date in the same query, and returns the top 10 matches to its drafting assistant.
Approximate search trades some accuracy for speed, so relevant results can be missed. A separate vector store adds data synchronization and duplicated permissions, and many workloads fit in an existing database with a vector extension.
Instead of matching exact values, the database ranks stored vectors by distance to a query vector using cosine similarity, Euclidean distance or dot product. Approximate nearest neighbor indexes such as HNSW graphs, inverted file (IVF) partitions, DiskANN or product quantization keep search fast over millions of Embeddings. Metadata filters can run before or after the vector search. Integrated options keep vectors next to source data, which simplifies Retrieval-Augmented Generation (RAG) pipelines.
Intelligent Automation, Systems Architecture
Pinecone — What is a Vector Database & How Does it Work? — https://www.pinecone.io/learn/vector-database/; Microsoft Learn — Vector database — https://learn.microsoft.com/en-us/azure/cosmos-db/vector-database