VyroAI

Loading...

AI Learning & Feedback

How VyroAI learns from your builds, and how 👍/👎 feedback improves it.

What is AI Learning?

VyroAI learns from what happens in the app: when a build succeeds it can extract useful patterns (event handler structures, command registration patterns) and store them in the knowledge base, and when a build fails it learns from the error and the fix that resolved it. Both make the AI better at the next one.

Your Code Studio generations count too. After a substantial turn there, your question and the AI's answer are both sent to be distilled into the same knowledge base, so it is not only builds and errors. Very short turns — a "yes" or "continue" — are skipped. If you would rather this did not happen, the switch below covers it.

Opting out

One switch controls all of it: Profile → Preferences → AI Assistant → "Learn from my projects". Turn it off and your builds, your errors, your Code Studio conversations and your feedback are all excluded from AI learning — see AI Assistant Preferences.

It is on by default. Turning it off does not delete anything already learned, and it does not change how the AI behaves for you.

Message feedback (👍/👎) — and it pays

Hover any AI message and click 👍 or 👎, optionally with a short comment. Recurring 👎 patterns drive prompt changes, model swaps and bug fixes.

Rating an answer can earn you credits: a few for a quick vote, more for a 👎 where you explain what went wrong. The reward has honest limits — once per message, at most 20 rewarded votes a day, and a lifetime cap: the rewarded votes on your account can never outnumber the billable generations you have actually completed. That last one is not a check on the individual message — it looks at your whole history, and works out to one paid vote per real run. See Earning Free Credits.

What is stored with it

Feedback is not anonymous — it is saved against your account and the project it came from, so the team can look at the actual conversation that went wrong, and it is processed as it arrives rather than in a weekly batch.

We never train models on your private projects. If you would rather none of this happened, the opt-out above covers it.

A 👎 with a one-line reason is 10x more useful than a 👎 alone. "Used deprecated API" or "Wrong MC version" tells us exactly what to fix.

Also in Advanced

3 of 4 in Advanced · Last checked 11 September 2026 · Previous: Sharing AI Instructions