Vector Database
Also known as: vector store, embedding database, ANN database
A vector database is a data store optimized for storing and searching high-dimensional vector embeddings, typically used to find semantically similar items for AI applications such as RAG, recommendation, and search.
Detailed explanation
Vector databases store numerical embeddings produced by ML models (e.g., text or image embeddings) and provide fast approximate-nearest-neighbor (ANN) search across millions or billions of vectors. They are a core component of RAG systems, semantic search, recommendation engines, and de-duplication pipelines.
Common engines include pgvector (Postgres extension), Pinecone, Weaviate, Qdrant, Milvus, and Chroma. Many traditional databases (Elastic, OpenSearch, MongoDB, Redis) now offer vector search as a feature. Choosing between a dedicated vector database and a vector-capable general database depends on scale, latency, hybrid-search needs, and operational complexity.
In production, vector search is usually combined with metadata filters and keyword search (hybrid retrieval) for accuracy. Reranking models often refine the top results before they are passed to the LLM.