Do I need to filter vector search by tenant and permissions?

Yes — in a multi-tenant or permissioned system, you should always filter vector search by tenant and access permissions.

Why

Vector search can return semantically similar matches from anywhere in the index. Without filtering, you risk:

• Cross-tenant data leakage
• Unauthorized document retrieval
• Incorrect or noisy results
• Compliance/security violations

Best practice

Use metadata filters alongside vector similarity search, for example:

• tenant_id = current_tenant
• access_level / acl / allowed_users / allowed_groups
• any other domain-specific constraints

Typical pattern

1. Narrow candidates with metadata filters
2. Run vector similarity search within that restricted set
3. Optionally rerank results

Important note

Permissions should be enforced at query time, not just at ingestion time. Even if you store separate indexes per tenant, still validate access control in the retrieval layer.

If you’re using RAG

This is especially important, because retrieved chunks may be shown to users or used to generate answers. A retrieval mistake can become a data exposure.

If you want, I can show you:

• a sample schema for tenant + ACL filtering,
• examples for Pinecone/Weaviate/Qdrant/FAISS,
• or a secure retrieval pattern for RAG.