AI You Can Check — Trust

How do you know an AI answer about your company is right?

You check it. Every answer carries the sources it came from — who said it, in which meeting or session, on which date — so any claim can be opened and verified. When there is no grounded answer, the system says so instead of guessing. And what counts as knowledge is decided by people: new knowledge waits for confirmation, and the expert keeps veto rights over his profile.

What does AI do in the approach?

AI does the heavy lifting that people should not have to do. It transcribes and structures conversations, extracts the statements that matter, and finds answers by meaning across everything captured. It can draft an instruction or a summary for review. It does not decide what is true for your company: people confirm, correct and decide.

Sources under every answer

Every statement in the knowledge base carries its origin: the speaker, the meeting or intake session, and the date. That is provenance. When an answer comes back, the sources come with it, so a disagreement is settled by opening the source, not by arguing about what the system meant. A summary counts for less than a person's own words, and is marked as such.

“I don't know” is an answer

Answers come from captured knowledge, not from what sounds plausible. When the knowledge base has nothing grounded to say, the answer says so. A digital knowledge copy always presents itself as a digital version, and it never invents opinions or memories that were not captured.

People confirm what counts

What AI extracts lands in a review queue first. It becomes trusted knowledge after a person confirms it, or when it clearly meets the agreed confidence threshold. The expert reviews what enters his profile and can strike anything. Corrections and feedback from colleagues go back into the same review, so the knowledge base improves where it was wrong.

Knowledge that changes over time

New information does not silently overwrite old information. Earlier versions keep their dates, so you can ask how something was done then and how it is done now — and see why it changed. This is what a temporal knowledge graph is for.

What we do not claim

A source proves where a statement came from, not that every interpretation of it is right. That is why people review.
An AI answer is not approved just because its sources are. Critical decisions stay with the people responsible.
No system knows what was never captured. Gaps show up as unanswered questions, and those are the next thing to capture.

Where MentX fits

MentX is a personality-aware knowledge base. It captures the knowledge of a company's key person — from meetings and a guided intake track — into a living, temporal knowledge graph with full source attribution. DigiME is the first use case built on the MentX knowledge base: a digital knowledge copy of your key person you can converse with. It answers in his knowledge, his reasoning style and his voice of thought. How that differs from a chatbot on your documents →

Can the AI make things up?

Any language model can produce a wrong sentence. The design limits the damage: answers must come from captured knowledge, sources are shown, “I don't know” is allowed, and people review what becomes trusted.

Is our knowledge used to train AI models?

No. Your knowledge is not pooled with other customers and not used to train anything outside your own Knowledge Space.

Who can see the sources?

Your company decides who may ask questions and who may open which sources. People who only ask questions do not get access to raw recordings unless you grant it.