Knowledge agents
you can defend.
Every answer is traceable to an approved source. Every source has an owner. Every owner has a verification stamp. And when the agent can't give a definitive answer, it refuses — visibly, measurably, and usefully.
This isn't another AI chatbot. This is a governance system for organizational knowledge.
The Problem Is Not Retrieval.
Any system can retrieve documents. Very few can tell you whether the answer is approved, current, owned, and defensible. When authority is unclear, teams hesitate, escalate, or guess. DIBRI removes that ambiguity — or stays silent.
Answer everything. Source nothing. No accountability. No way to know if the answer is current, approved, or fabricated.
Search documents and generate answers. But who approved the content? Is it current? Who's accountable when it's wrong?
Every answer is verified, owned, sourced to exact passages, and time-bound. When it can't answer definitively, it refuses — and that refusal becomes data.
Four Trust Pillars.
Every answer must pass four gates. If any gate fails, the agent refuses.
Scoped
Agents answer only within assigned vaults. They cannot cross authority boundaries, speculate beyond their scope, or access content they haven't been granted.
- ✓Each agent tied to one vault — clear, enforceable scope
- ✓Refusal analytics — see when and why agents refuse
- ✓Knowledge gap radar — refused questions become a content backlog
Sourced
Every answer is traceable to exact documents and passages — not just a filename, but the specific paragraph the agent used. Click any citation to verify.
- ✓Deep citations — click [1] to see the exact passage used
- ✓Confidence scoring — 0–100 certainty signal on every answer
- ✓Document grounding notes — owner-set rules shape how evidence is interpreted
Current
Vaults represent the state of truth today — not last quarter. Stale content is detected, flagged, and blocked from powering answers until refreshed.
- ✓Freshness gates — block answers when documents exceed their review window
- ✓Staleness alerts — daily detection with owner notifications
- ✓Scheduled crawling — auto-refresh on configurable intervals
- ✓Version history — SHA-256 tracked changes on every re-ingestion
Owned
Every vault has a named human owner who is accountable for the knowledge within it. Content doesn't become an answer until the owner stamps it as verified.
- ✓Vault ownership — named accountable party per knowledge domain
- ✓Verification stamps — owner must approve before content powers answers
- ✓Feedback loop — end users flag wrong answers, routed to vault owner
How It Works.
From raw content to verified, defensible answers in four steps.
Create a Vault with an Owner
A vault isn't a folder — it's a governance boundary. Every vault has a named owner who is accountable for the knowledge within it.
- ›Set a freshness gate — e.g. 90 days. If all documents are older, the agent refuses to answer.
- ›Enable content approval — new uploads require the owner's verification stamp.
- ›Assign groups to control who can see and query the vault.
Ingest Evidence, Not Answers
Content enters the vault as evidence. It does not become an answer until the vault owner reviews it, optionally adds grounding notes, and stamps it as verified.
- ›Anyone can suggest content — regular users upload files for review
- ›Owners preview the actual content before approving — see every chunk
- ›Add grounding notes — rules like "superseded policy — cite for historical context only"
- ›Content changes tracked with SHA-256 hashing and version history
Questions Are Evaluated, Not Just Answered
When a question is asked, DIBRI doesn't just search for relevant text. It evaluates whether the vault authorizes an answer.
Two gates must pass before any answer is rendered:
- ›Sourcing gate — are there approved, relevant documents for this query?
- ›Freshness gate — are those documents within the vault's review window?
If either gate fails, the agent refuses — and the refusal is logged, tracked, and surfaced to admins as a measurable signal.
Answers with Proof
When the vault does authorize an answer, you get more than text. You get verifiable proof.
- ›Deep citations — click any [1] to see the exact passage, not just the document name
- ›Confidence score — a 0–100 signal based on source quality, diversity, and match strength
- ›Feedback loop — users can rate answers and flag incorrect responses, routed directly to the vault owner
The Trust Dashboard.
Complete visibility into whether your knowledge agents are keeping their promises.
Refusal Analytics
See how often agents refuse, why, and for which vaults. Every refusal is logged with the question, gate type, and timestamp.
Staleness Alerts
Daily checks flag documents past their review window. Color-coded severity with one-click acknowledgment for vault owners.
Knowledge Gap Radar
Clusters refused questions by topic and ranks them by frequency. Your content backlog writes itself — build what users actually need.
Version History
Every re-ingestion is tracked with content hashing. See what changed, when, and whether content is actually different — not just re-uploaded.
Flagged Answers
End users flag incorrect answers with one click. Flags route to the vault owner with the original query, response, and reason — close the loop.
Document Quality Scores
Aggregate user feedback per document. See which content drives positive outcomes and which needs attention — data-driven content quality.
Agents Don't Decide. Vaults Do.
Agents are scoped to vaults. They cannot speculate, override governance, or cross authority boundaries. They only render what the vault permits — and when they can't, they refuse, and that refusal creates signal.
Deploy Anywhere
Embed agents on your website, share public links, or use internally. Each agent inherits its vault's governance rules.
One Agent, One Vault
Each agent is bound to exactly one vault — ensuring clear authority boundaries. No cross-contamination, no ambiguous sourcing.
User Feedback Built In
Every response includes thumbs up/down and flag-as-incorrect. Feedback is linked to the cited documents and routed to the vault owner.
Teams That Need Answers They Can Stand Behind.
No wrong answers, no gray areas, no ambiguity.
Healthcare & Life Sciences
Clinical protocols, drug information, and internal guidelines — served only from approved, current sources with full traceability.
Financial Services
Policy manuals, internal documentation, and procedures — with version tracking and accountability at every step.
Legal
Contracts, guidance documents, and reference materials — scoped vaults ensure matter-level isolation.
Government & Public Sector
Policy documents, public-facing FAQs, and internal procedures with ownership and approval controls.
Customer Support
Deploy agents that answer from your approved knowledge base — with citations, not hallucinations. Agents refuse when they don't know.
Professional Services
Consulting, accounting, and advisory firms — deliver verified answers to clients with source tracing and vault-level isolation per engagement.
If It Comes From DIBRI,
You Can Defend It.
Definitive answers. Verified sources. Named owners. Measurable trust.
Build authority into your knowledge — and let AI operate within it.