Multi-tenancy is an architecture in which a single vector database instance serves many separate customers, or tenants, while keeping each tenant’s data isolated from the others. It lets a provider run one shared system to support countless customers, rather than standing up separate infrastructure for each.
The central challenge is balancing isolation, performance, and cost. Tenants must not see each other’s data, and ideally one tenant’s heavy usage should not slow down another — the noisy-neighbour problem. Yet running fully separate systems per tenant is expensive, so multi-tenant designs share infrastructure while enforcing boundaries through one of several patterns.
Those patterns span a spectrum: a shared collection with a tenant-ID filter, a namespace per tenant, a collection per tenant, or a fully separate database per tenant. They trade off isolation strength against efficiency — lightweight logical separation scales to many small tenants cheaply, while physical separation gives strong guarantees for large or sensitive tenants. Strong multi-tenancy support, including features like per-tenant shards and tenant lifecycle management, is a key differentiator for vector databases serving SaaS applications.