Your Complete Guide

How to Get the Most from DIBRI

Build trustworthy AI knowledge agents that answer only from your authoritative sources. Every response is scoped, sourced, current, and owned.

This guide walks you through everything DIBRI can do — from your first vault to a fully governed knowledge system.

The Four Trust Pillars

Every DIBRI agent is built on four pillars that ensure your AI knowledge system is defensible and accountable.

Scoped

Agents answer only from documents in their vault. Nothing else enters the response.

Sourced

Every answer includes deep citations — click any reference to see the exact source passage.

Current

Staleness alerts and scheduled crawls ensure your knowledge base never falls behind.

Owned

Every vault has an accountable owner. Every document has an approval stamp and audit trail.

How It Works

From zero to a trustworthy AI knowledge agent in four steps.

1

Create a Knowledge Vault

Upload your authoritative documents — PDFs, Word files, web pages. Each vault is a self-contained knowledge boundary.

2

Deploy an AI Agent

One agent per vault. It answers only from your sources — never from the internet, never from training data, never from guessing.

3

Monitor Trust Health

Real-time dashboards show confidence scores, refusal analytics, knowledge gaps, and flagged answers — all in one place.

4

Keep Knowledge Current

Scheduled crawls auto-refresh your sources. Staleness alerts notify you when content falls behind. Version history tracks every change.

What You Can Do

A full toolkit for building, governing, and deploying trustworthy AI knowledge agents.

Upload & Crawl

Drag and drop files, crawl entire websites, or sync from OneDrive. All content is chunked and indexed automatically.

Evidence Approval

Require human review before any content goes live. Only vault owners can approve — with a verification stamp and audit trail.

Confidence Scoring

Every response is rated 0-100 using four signals: distance, passages, sources, and grounding. Displayed as a green/yellow/red bar.

Deep Citations

Click any [1] reference in a response to see the exact source passage. Chunk-level precision, not just filename-level.

Refusal Analytics

See every question your agent chose not to answer. Refusals become data — showing where your knowledge has gaps.

Knowledge Gap Radar

Discover what content your users are looking for but not finding. Clustered keyword analysis reveals content opportunities.

Scheduled Crawling

Auto-refresh your content on a schedule — hourly, daily, or weekly. Never worry about stale information again.

Version History

Every document change is SHA-256 tracked. See exactly when content changed, what changed, and who approved it.

API & Widgets

REST API for integrations, an embeddable chat widget for your website, and shareable direct links for public access.

Trust Dashboard Preview

Eight trust views give you complete visibility into how your AI agent is performing and where to improve.

Refusals
Staleness
Knowledge Gaps
Flagged Answers
Versions
Confidence
Documents
Approvals

Refusal Analytics

Time-series chart of every question your agent refused. Filter by gate type, vault, and date range. Spot trends before they become problems.

Knowledge Gap Radar

Clustered keyword analysis from refused queries. See what your users need but your vault doesn't have. Prioritize content creation with data.

Staleness Alerts

Per-document and per-vault age tracking. Automatic alerts when content exceeds your freshness threshold. Never serve outdated answers.

Flagged Answers

Users can flag incorrect answers with a reason. Admins review, resolve, and improve. A closed-loop feedback system for answer quality.

Agent Gate Types

When your agent can't give a trustworthy answer, it activates one of these gates — and logs the event for your dashboard.

Scope Gate — query outside vault boundary Confidence Gate — score below threshold Fallback Gate — no matching documents found Speculation Gate — answer not grounded in sources

Deployment Options

Three ways to put your AI knowledge agent in front of users — all with the same trust guarantees.

REST API

Full programmatic access with token-based authentication. Stream or batch responses with confidence scores and citations.

POST /api/v1/agent/chat
Authorization: Bearer <token>

Use case: Internal tools, Slack bots, CRM integrations, custom dashboards.

Chat Widget

Embed a branded chat widget on any webpage with a single script tag. Customizable colors, welcome message, and behavior.

<script src="/widget.js"
  data-agent="your-token">

Use case: Customer support on your website, product documentation help.

Direct Link

Share a standalone chat page with anyone. No login required for public agents. Full trust features including citations and confidence.

dibri.app/agent/your-agent-slug

Use case: Public-facing knowledge bases, partner portals, onboarding guides.

Ready to build trust into your AI?

Start with a free trial. No credit card required. Your first knowledge vault is live in minutes.