Upstash Vector: how to partition data per tenant (namespaces/multi-tenancy)

Upstash Vector: how to partition data per tenant (namespaces/multi-tenancy)

The problem

A user building multi-tenant RAG on Upstash Vector asked for the equivalent of namespaces or partition keys used by other vector providers. Their users upload their own documents and must only be able to query their own. At the time, Upstash had no namespace concept, so the user was looking for any supported way to partition the data.

The verified fix

At the time the answer was metadata filtering: Upstash shipped SQL-style metadata filtering (see https://upstash.com/docs/vector/features/filtering), so you add a tenant field to each vector's metadata and filter on it per query. Namespaces were planned for a later stage. Source: https://github.com/upstash/vector-js/issues/12

Asked in the thread (https://vectle.com/threads/thrCbr42DPUAKhoe2yz0xCtFg): A user building multi-tenant RAG on Upstash Vector asked for the equivalent of namespaces or partition keys used by other vector providers. Their users upload their own documents and must only be able to query their own. At the time, Upstash had no namespace concept, so the user was looking for any supported way to partition the data. The verified answer: At the time the answer was metadata filtering: Upstash shipped SQL-style metadata filtering (see https://upstash.com/docs/vector/features/filtering), so you add a tenant field to each vector's metadata and filter on it per query. Namespaces were planned for a later stage. Source: https://github.com/upstash/vector-js/issues/12 Thread: https://vectle.com/threads/thrCbr42DPUAKhoe2yz0xCtFg