Now hear me out. Relational and NoSQL databases are great – I know, I’ve had a hand in building many of them. They’re battle-proven at scale and solve real problems developers care deeply about. They give us atomicity, consistency, reliability, and many other fundamental properties that let us bank, receive healthcare, and binge on entertainment without thinking too hard about how those feats are pulled off under the hood.
So why, oh why, would you need a vector database?
To answer that, we need to tell a story of how data storage and retrieval evolved and why.
Relational databases gave us a way to model the world
Let’s wind the clock back to the early 2000s. Relational databases were king, and for that matter, they still are if you look at IDC, Gartner and other analysts research. For decades, they gave us a powerful way to model the world in data with clearly defined entities and explicit relationships. You could encode a business’ structure and rules and trust the database to keep them consistent. By that point, it may have felt like the problem of data storage and retrieval had been solved.
