mediumAI Engineering

How do you scope retrieval with metadata filtering and multi-tenancy?

Q: How do you scope retrieval with metadata filtering and multi-tenancy?

**TL;DR.** Store **tenant id, ACLs and attributes** beside each vector and apply them as **pre-filters inside the ANN query** — users can only retrieve chunks they are authorized to see. Access control lives in **retrieval**, never in the prompt ("do not reveal other tenants" is not a security boundary). query tenant=42, role=hr filtered ANN <text class='d-sub' x='233'

322 views

Understand the problem

Pre-filtering vectors by tenant, permission and attributes — where access control lives in RAG.

metadata-filteringmulti-tenancysecurityrag

Attempt it yourself

Sketch your approach before reading the solution — that's what interviews test.

Stuck? AI Nudge Available

Get a conceptual hint to guide your logic without spoiling the final implementation.

Study the solution

The solution is waiting

Give it an honest attempt first — then compare your thinking with the full walkthrough.

Read the code

Server-derived filters in the retrieval call

async function retrieveForUser(query: string, session: Session, k = 8) {
  const qVec = await embed(query);
  return vectors.search(qVec, {
    limit: k,
    filter: {
      tenant_id: session.tenantId,          // from the session, NEVER from input
      acl: { any_of: session.groups },
      status: "published",
    },
  });
}
// deleting a doc / revoking access must also remove or reflag its chunks

Join the discussion

Discussion (0)

No responses yet. Be the first to share what you think.

What is LLM-as-judge and what are its known biases?

How do you handle PII and sensitive data when calling LLM APIs?