AI Chat Assistant

Legal Engineering with AI

An assistant that only answers what you gave it.

Embeddable AI chat that queries your own knowledge base instead of improvising. Built for operations where a wrong answer has a cost — precision, traceability and AI disclosure by design.

How it works

Not a generic chatbot

Most "ready to use" widgets either do not know your real content, or freely mix it with the model's generic knowledge. This one does not.

Answers from your documents, not the internet

RAG over your own knowledge base — policies, processes, template contracts, internal FAQs.

Traceability, not hallucination

The "direct article" shortcut answers by reconstructing the original text from your knowledge base, skipping the model when it is not needed.

Isolation and disclosure by design

Per-client RLS, rate limiting, manipulation detection, and a fixed "you are talking to AI" notice that does not depend on editable configuration.

Knows when it needs a human

Classifies the reason for contact and routes it to the right person at your firm — the category-to-person mapping is fully editable from your panel.

Sound familiar?

When it makes sense

Your team answers the same repeated questions over email or WhatsApp
You have internal documentation nobody external can query directly
The generic chatbots you tried invent answers outside your content
You need an auditable record of what each visitor asked
Requests that need human judgment get lost in a generic inbox nobody owns
Validated in production

Already running at Clever Management

The same engine answers real visitor questions today about the agency, courses, the scouting process and contact for Clever Management — with real-time streaming, cached answers for frequent questions, and the direct-article shortcut that skips the model.

FAQ

Frequently Asked Questions

Can it make up information that is not in my documents?

The design prioritizes answering from your knowledge base (RAG). For queries pointing to a specific article, a shortcut reconstructs the original text from your documents without going through the model — the most traceable source possible.

Does my data train the model?

No. Your knowledge base is used as query context (RAG), not as training data.

Does it comply with mandatory AI disclosure?

Yes — the widget always shows a notice that the visitor is talking to an automated assistant, independent from the rest of the editable configuration.

How is it installed?

One script line on your site. Mode (floating button or fullpage), visual style and brand attribution are all configurable without touching code.

What happens when a request needs a real person?

The assistant detects the reason for contact (sales, support, billing, etc.) and escalates it to the right person or team, defined by you from the panel. There is a record of who each request was routed to and whether the notification was delivered, so nothing falls through the cracks.

Add an assistant that does not improvise to your site

Tell us about your operation and we will show you how it installs.