VaultLens

VaultLens

Permission-aware document intelligence for trustworthy enterprise AI architecture.

VaultLens demonstrates how company-document Q&A systems can enforce role-based access, cite supporting sources, block restricted content, log audit events, and evaluate answer behavior — using a controlled synthetic document set. This is a RAG-ready trust-layer prototype, not a live retrieval-augmented generative pipeline.

Ask questions across synthetic company documents with role-aware access checks, source citations, audit logs, and evaluation cases.

Try DemoRead Case Study

How it flows

  1. Role
  2. Question
  3. Access filter
  4. Retrieval
  5. Cited answer
  6. Audit log

Why it exists

Enterprise teams need answers that are grounded, permission-safe, and reviewable. This demo shows how to enforce access before quoting text — and how to test that behavior.

What you get

  • • Role-based allow lists (demo switcher, not SSO)
  • • Citations for every grounded answer
  • • Explicit handling when the best match is off-limits
  • • Audit-style event bundle + evaluation dashboard

Routes

Public demo. This site uses deterministic retrieval over synthetic documents. It does not call a live generative model or embedding API. The architecture is RAG-ready: Supabase/Postgres, pgvector, and an LLM can be added later while keeping the same access-control, citation, audit, and evaluation layers.

Responsible use (demo)

  • Public demo uses synthetic documents only. Do not upload private files in v1.
  • Restricted documents are never revealed to roles that cannot access them.
  • Grounded answers require cited sources; support scores are retrieval heuristics, not guarantees.
  • When in doubt, have a subject-matter expert review the answer in production.