Edtek Chat
Turn your content into an intelligent chatbot.
Source-verified AI chatbot that answers queries from your trusted documents — with page-level citations, not hallucinations.
What it does
Answers a careful reader can check.
Three things happen on every question. Together they are the difference between a chatbot over your documents and a chatbot that has read about your subject.
Grounded answers
Every claim opens to the page it came from
Citation numbers are handed out by the retrieval step, not written by the model, and a reference to a passage that was not retrieved never reaches the screen. Clicking one opens the source next to the answer and paints the exact passage.
- Page-level references in PDF; Word and Markdown highlight the cited block
- The sources card separates what was cited from what was found and not used
- Expand any source to read the retrieved passage and its match score
- The document panel is a real viewer — page navigation, jump-to-page, zoom
Honest failure
When your library has no answer, it says so
If nothing matches closely enough, there is no grounded answer to give — and the reader is told that, plus what would change the outcome, instead of being handed something plausible.
- On a first question, no matching passages means no model call at all
- Mid-conversation, it may answer from the conversation so far — never from outside knowledge
- Answers built on weaker matches are labelled, not hidden
- Where the sources only partly cover a question, the answer says which part
Your product
One installation, as many chatbots as you need
An instance is a chatbot: its own name, mark, palette, welcome screen, prompt, model, retrieval tuning, members and shelf. All of it configured in an admin console.
- Six themes with light and dark treatments, plus two densities
- Settings resolve instance → global → shipped default, with the source shown on every field
- Editable system prompt, with the shipped default one click away
- Per-deployment collection allowlists, enforced before any search runs
Measured, not asserted
Answer quality has a number.
Edtek Chat ships with a regression suite for its own answers. Golden questions with reference answers, an LLM judge, assertions about what retrieval had to return, and latency and cost budgets — all executed through the same pipeline that serves your readers.
Mark a run as the baseline, change a model or a threshold, and get a list of what regressed before your readers find out.
How evaluation worksWho uses Edtek Chat
Arbitration & dispute resolution
Practitioners query case preparation guides, rules, and procedural handbooks in plain English — and get answers with exact page citations.
Academic & reference publishers
Turn backlist titles, encyclopedias, and research archives into interactive Chat Books. Readers discover answers they'd never find by browsing.
Compliance & regulatory teams
Staff query internal policies, regulatory frameworks, and procedural manuals instantly — no more hunting through SharePoint or email threads.
How Edtek Chat works
From documents to conversational intelligence in days, not months.
- 1
Load your library
PDFs, Word files, Markdown and plain text go into collections you organise. Files and extracted text stay in storage you control.
- 2
It is indexed, passage by passage
Documents are split without breaking sentences, embedded with the model you choose, and stored in your own vector index.
- 3
Readers ask in plain language
No syntax, no training. They can narrow a question to a collection or to single titles before asking.
- 4
Answers arrive cited
Each claim carries a numbered reference that opens the source beside the answer, with the retrieved passage highlighted.
Capabilities
Citations that open the page
A numbered marker on every claim opens the source beside the answer, jumps to the page and highlights the passage it came from.
A refusal, not a guess
When nothing in the selected sources matches, there is no grounded answer to fall back on. The reader is told, and told what to try instead.
Scope before asking
Readers narrow a question to a collection or to individual titles, and each deployment starts from a default scope you set.
Your brand, per deployment
Name, mark, palette, density, welcome copy and tips are settings. Six themes, light and dark, optionally handed to the reader.
Your models, your keys
OpenAI, Azure OpenAI, OpenRouter, Cohere for reranking, or any OpenAI-compatible endpoint — under credentials encrypted at rest in your own deployment.
A test suite for answers
Golden questions, an LLM judge, retrieval assertions and latency budgets, with baselines and scheduled runs to catch regressions.
Go deeper
Four things worth reading before you book anything.
Grounded answers
Citations that open to the page, and what happens when there is no answer.
Read moreCustomization
Instances, branding, prompts and retrieval — configured, not coded.
Read moreEvaluation
A test suite for answer quality, with regressions caught before release.
Read moreEnterprise
Your infrastructure, your models, your tenants, your audit trail.
Read more
Featured case study
AAA: The world's first AI-powered arbitration Chat Book.
“The AAAi Chat Book allows neutrals and practitioners to query over a century of arbitration wisdom and instant access to specific rules with full source verification.”
American Arbitration AssociationLaunched Jan 2025
Frequently asked questions
What is Edtek Chat?
Edtek Chat is a source-verified AI chatbot that answers queries exclusively from your uploaded documents. Every response includes page-level citations so users can verify answers against the original source material.
How does Edtek Chat prevent hallucinations?
Answers are constrained to passages retrieved from your library, and the constraint is enforced in code rather than only requested in a prompt: citation numbers are assigned by the retrieval step and any reference to a passage that was not retrieved is stripped before the answer is shown; retrieved passages always reach the prompt even if a custom prompt forgets them; and when nothing in the selected sources matches on a first question, the answering model is never called at all — the reader is told the library has no answer. Mid-conversation the model may instead be asked to answer from the conversation so far, under a prompt that forbids outside knowledge. Optional relevance and verification passes add a second model that filters passages before the answer and reviews the draft against its citations afterwards.
What types of documents can I upload?
PDF, Word, Markdown and plain text, organised into nested collections. Sources open in a reading panel beside the answer with the cited passage highlighted — PDFs are the only format with real pages, so they alone carry a page reference; Word and Markdown highlight the cited block.
Can I deploy Edtek Chat on my own infrastructure?
Yes. Edtek Chat supports on-premise deployment for organizations with strict data residency or compliance requirements. Your content never leaves your infrastructure.
Can we run more than one chatbot?
Yes — as many as you need from one installation. Each is an instance with its own branding, welcome screen, prompt, model, retrieval tuning, members and shelf of collections, at its own URL. They are configured in an admin console, not built separately.
How long does it take to set up?
Most deployments go from document upload to live chatbot in days, not months. No programming or prompt engineering is required — you describe your chatbot in plain English.
One platform, three products
Connected intelligence across your entire content workflow.
Edtek Draft
Document Builder
Create templates from your documents and generate at scale, or connect a database and let AI verify, rewrite, and enhance every section.
Learn moreEdtek Cite
Smart Annotations
Surface the rules, cases, and authorities behind every claim — inline, with context and summaries.
Learn moreSee edtek.ai in action.
Book a 30-minute demo with our team to explore how our products can transform your organization's verified content into intelligent tools.