A database designed to store embeddings and find similar ones fast.
A vector database is a special kind of database. Instead of letting you search by exact match ('find rows where email = bob@...'), it lets you search by similarity ('find the embeddings closest to this one'). It can do this across millions or billions of items in milliseconds because it uses a clever index tuned for high-dimensional number comparisons.
Without one, RAG doesn't scale. You can fake it with a Python list for 100 documents. You can't for 100,000.