Once you have embeddings, you need somewhere to store them, search them, filter them, update them, and keep them connected to the source records. That is the job of a vector database.

A vector database retrieves records by similarity rather than exact equality. Given a query vector, it returns nearby vectors under a metric such as cosine distance, dot product, or Euclidean distance. In real systems, it also applies metadata filters, enforces tenant boundaries, supports deletes and updates, provides operational visibility, and survives restarts.

This chapter focuses on the engineering trade-offs behind those systems: why brute force stops being enough, how approximate nearest neighbor indexes work, and how to choose between a dedicated vector database and vector search inside an existing database.