In plain English
A vector database is a specialised store designed to hold embeddings and search them efficiently by similarity. Instead of matching exact keywords, it finds the items whose meaning is closest to a query. It is a core piece of infrastructure for retrieval-augmented generation and semantic search.
Why it matters
Vector databases are what let an AI assistant pull the most relevant passages from your documents by meaning, which is the retrieval half of most RAG systems.
A worked example
A search for 'how to reset my password' returns a help article titled 'recovering account access', because their embeddings sit close together even with no shared words.
Common confusion
A vector database stores numerical embeddings, not your original documents, and it is not a replacement for a normal database for exact lookups like an order number.