Tenant isolation is the separation of one customer’s data and query workload from another’s in a multi-tenant vector database, so that tenants cannot see each other’s data and ideally do not interfere with each other’s performance. It is the central requirement that makes it safe to serve many customers from shared infrastructure.
Isolation operates on two dimensions: data and performance. Data isolation ensures a tenant’s queries only ever return that tenant’s records, enforced through namespaces, partitions, per-tenant collections, or row-level security. Performance isolation ensures one tenant’s heavy usage does not degrade others — the noisy-neighbour concern — which is harder to achieve when resources are shared.
Approaches range along a spectrum from logical to physical. Logical isolation shares infrastructure and enforces separation in software, which is efficient and scales to many tenants. Physical isolation gives each tenant dedicated resources for the strongest guarantees at higher cost. Choosing the right model — and features like per-tenant shards and tenant lifecycle management — is a key part of designing a multi-tenant SaaS application on a vector database.