Vector Databases

Chapter 1. Introduction to Vector Databases

The current wave of generative AI (GenAI for short), especially in natural language applications, has brought the technology of vector databases to the forefront of developer attention. Unstructured data used to train models is vectorized, and data used in inference is often saved in vector databases. And yet the technology is not well understood or widely known to the average database application developer.

This chapter provides a high-level overview of vector databases, outlining their main components and offering a preview of the material ahead. Mostly, you’ll learn what the technology is and why you need it. By the end of the chapter, you’ll understand how to use vector databases; the key topic of embeddings will be covered in the next chapter.

Why Do You Need Vector Databases?

Even before the advent of GenAI, managing unstructured data such as text, images, audio, and video was difficult. It required specialized technology and knowledge to manage these assets at scale. Indexing and searching over such data were only possible at the level of the metadata—for example, the date of creation, size, author, and so on. But searching for an image similar to another one, or a paragraph with a similar meaning to another, was not something an SQL database could do. SQL databases are used to match on syntax, not semantics.

Now, with the emergence of vector databases into the mainstream, managing unstructured data and searching based on semantics ...