How we train

Trained in the open.

Ultrec learns from work it is allowed to keep, inside a daily limit. Here is the plain version. No benchmark theater.

Notes it may keep

A saved note is something a person chose to leave for next time. It is not a scrape of a private drive.

Skills with a source

A skill is a small, named way of working. If we cannot point at where it came from, it does not ship.

A daily limit

Practice stops at the cap. More hours are not a goal. The cap is there so a helper cannot run on and on.

People still decide

Training does not get a vote over the safety rules. Sandbox, hidden keys, and the cap stay on.

Where the line is

What we will not do to “get smarter.”

  • We don’t quietly widen what a tool can touch so a model can see more.
  • We don’t train on material a person did not mean to hand over.
  • We don’t publish a score we cannot show the work behind.
  • We don’t call a draft a finished method.

If we cannot show the work, we do not quote a number.

Ultrec AI Inc · how we train

No benchmark scores on this page. If we cannot show the work, we do not quote a number.