What is an embedded vector database?

An embedded database is one that runs as a library directly inside your application's own process, rather than as a separate server your application connects to over a network — the same architectural pattern SQLite uses for relational data, applied here to vector search.

Because there's no network hop between the application and the database, an embedded vector database like LanceDB avoids the serialization and round-trip latency that comes with a client-server architecture, and it drastically simplifies deployment: there's no separate database process to provision, scale, or keep running alongside the application.

The trade-off is that a purely embedded database is naturally suited to single-process access; scenarios needing many separate processes or machines to concurrently read and write the same data typically call for either careful use of shared object storage or a managed cloud offering built on the same underlying format, which is exactly the gap LanceDB Cloud is designed to fill.