medium10 pts · first answer countsragsecurity

Keeping tenants apart in RAG

Your RAG SaaS stores many customers documents in one vector index. How should you stop one customer from seeing another customer content?

A.Tell the model in the system prompt to only use the current customer documentsB.Retrieve the top 100 results, then ask the model to drop other tenants chunksC.Apply a tenant ID metadata filter inside the vector query, enforced server sideD.Use a higher similarity threshold so other tenants rarely match

← Counting chunks with overlap Tuning HNSW search breadth →