Vector Databases

Definition: A vector database stores embeddings and finds the most similar ones to a query extremely fast — even across millions of vectors. It is the "memory" that makes retrieval practical at scale.

What it does

You give it vectors (plus the original text). At query time you give it a question vector, and it returns the top-K most similar chunks in milliseconds. Doing this by hand — comparing against every vector — would be far too slow at scale, so vector databases use clever indexing.

Popular options

• Pinecone, Weaviate, Qdrant, Milvus — dedicated vector databases
• FAISS, Chroma — libraries you can run locally
• pgvector — adds vectors to PostgreSQL (great if you already use it)

What to know as a beginner

They all do the same core job: store vectors, search by similarity, return the closest chunks. Pick based on scale, cost, and whether you want a managed service or something self-hosted.

💡 Mental model: a vector database is just a search engine that searches by meaning instead of by keywords.