The need for vector databases

Vectors carry deep semantic information and have many potential use cases that will make them increasingly common over the next few years. Working with them requires specific and complex operations that only process vector data. Additionally, the demand for search can often vary substantially from the demand for more structured database queries.

Together, these factors mean vector operations and traditional database workloads are largely independent. This gives rise to the concept of a vector database that’s designed specifically to handle vector data, indexes, and workloads. From a developer’s perspective, vector databases can take several forms.

The most basic is a standalone product that’s independent from other operational databases. This type of vector database has the freedom to focus solely on implementing and optimizing vector operations without considering other database operations. However, often, vector search applications...