How it works
Three steps. Zero effort.
Correct
Just do what you already do: edit AI output.
brain.correct(
draft="Dear Sir...",
final="Hey..."
)Learn
Gradata figures out what you meant and remembers it.
Correction detected
"Dear Sir..." → "Hey..."
Rule graduated
No em dashes
Never again
Your AI applies what it learned to every future session. Automatically.
# System prompt (auto-injected)
Apply learned user preferences before responding.
Learned rule applied:
No em dashes
Rules need repeated signals before they stick. One bad correction can't break anything.
Try it live
See for yourself. No sign-up needed.
Watch a rule graduate from scratch. Three corrections, one pattern, permanent memory.
Submit the correction above to see the brain learn.
Trust
What if it learns the wrong thing?
Mistakes don't stick.
It takes multiple similar corrections before anything becomes a rule. One bad edit, one typo, one off day: the brain ignores it. Only consistent patterns stick.
It self-corrects.
If a learned rule starts making output worse, the brain detects the drop and turns it off automatically. No babysitting required.
You're always in control.
See something you don't like? Undo it. One line. The rule is gone.
brain.forget("casual tone")