Definition
Vector Database
A vector database stores embeddings and answers nearest-neighbor queries: given a query vector, return the stored chunks closest to it in meaning. It is the memory layer of RAG. Pinecone, Weaviate, Qdrant, and ChromaDB are the common choices; FAISS works for prototyping. Chunk size matters, with 500 to 1,000 tokens as a typical default, because chunks too large lose precision and too small lose context.
Explained in
Chapter 9: What Is an Agent?
The moment you give an LLM access to knowledge or tools, it becomes an agent. And that makes all the difference.
Related terms