Vector-based retrieval has become a foundation piece of generative AI. From an engineering perspective vector search differs from traditional search which led to the uprising of new vector-native databases and vector search integrations into existing databases.

When AI startups and AI-native larger companies use vector search they typically have multi-tenancy cases; they have users that have data that needs to be isolated from that of other users. In the recent past it was only possible to solve this with workarounds: Using filtered vector search or isolating tenants by putting them in dedicated collections or namespaces. But those approaches were full of disadvantages and didn’t scale to millions of tenants with billions of vectors. To solve multi-tenancy in vector search – and enable the generative AI startup of the future – a paradigm shift is needed.