Skip to content

Every claim is checked before you rely on it

Concepts

The risk with AI is not just being wrong. It is being wrong confidently. Telha reduces that risk by generating from evidence first, then checking each statement against that evidence before presenting the answer as supported.

Grounded, then checked

When you ask a question, Telha does two things most assistants skip:

  1. It gathers the real evidence first. It pulls the relevant facts and their sources out of your connected systems before writing anything, and writes the answer from that evidence only, not from general knowledge.
  2. It checks the answer against that evidence. It breaks the answer into individual statements and confirms each one is actually supported by the sources it used.

Supported, partial, or unsupported

Each statement in an answer gets a verdict, so you know exactly which parts you can stand behind:

Verdict Meaning What to do
Supported The sources back this up. Safe to use in an answer, brief, or response.
Partial The sources back part of it. Review the caveat before relying on it.
Unsupported The sources do not back this. Do not use it as fact; Telha flags it rather than presenting it.

You see this as the "verified" line and citations on an answer. It means you are never left guessing which parts to trust, which is what makes an answer safe to put in front of a regulator, insurer, auditor, or board.

When there is nothing to stand on

If Telha cannot find evidence for your question, it tells you it does not have enough to answer, rather than inventing something. That "I don't know, and here's why" is a feature: it is how you know the "yes" answers are real, and it is how a brief shows you where the evidence is still missing.

Why this matters to you

  • Confidence you can act on. A verified answer is one you can put in a board pack, a claim response, or a regulatory submission.
  • No silent mistakes. Anything shaky is flagged, not hidden.
  • A clear trail. Every verdict traces back to a source you can open.