How Vector Search works

In this lesson, you will get a high-level understanding of how Vector Search is implemented in ScyllaDB.

ScyllaDB separates vector indexing from the core database by using a dedicated Rust-based Vector Store. Each ScyllaDB replica pairs with a local Vector Store node in the same availability zone.

ScyllaDB stores tables with vectors, and the Vector Store builds indexes from this data using CQL and CDC. When a client queries ScyllaDB, it retrieves the top neighbors from the Vector Store via HTTP.

The core of our Vector Store is built on top of the USearch engine. We also use a set of Rust services to interface with ScyllaDB, build vector indexes, and provide search capabilities.

For optimal performance, the Vector Store keeps all indexes in memory. This means that the entire index needs to fit into a single node’s RAM. We’re exploring hybrid approaches for future iterations.

To learn more about our implementation and benchmarks, read the blog post.