VyroAI

Loading...

Auto-Fix

When a build fails, the AI reads the compiler output and patches every file involved. How the repair rounds work, and when they cost credits.

How auto-fix works

When a build fails, VyroAI reads the compiler output, works out what actually caused it, and patches every file it needs to in one go — it deliberately does not stop after the first one, because a single root cause often breaks several files at once. If the first attempt does not work, it analyses deeper before trying again.

A Build run checks its own compile first — and can bill repair rounds

Since 10 September 2026, a Build run on a plugin or mod project does not hand over as soon as the files are written. Our server compiles the project first, and the chat shows Running a build check — compiling your project on our build server…. If the compile fails, the AI gets up to two repair rounds, and the project is compiled again after each one. Once it compiles, it can get one more round for problems the checks after the build found. The run ends with one Build check line: that the project compiles, what is still wrong, or Build check skipped with the reason.

Those rounds belong to the same run, and every round whose change is kept is billed on that run. So one Build run can carry up to three extra rounds on top of writing the files. A round that changes nothing costs nothing. If the repair leaves the project with more errors than it started with, your files are put back exactly as they were, and that attempt is not charged. Pressing Auto-Fix in Build mode starts a Build run too, so it gets the same check and can carry the same rounds.

The two switches in the Build & Export menu do not stop this. Auto-compile after AI and Stop on error (no auto-fix) only decide what the app does after a build: with Stop on error on, the app will not press Auto-Fix for you. The repair rounds inside a Build run happen on our server either way, and there is no setting on your side that turns them off.

How to trigger it

The AI shows what it changed and why.

Auto-Fix spends credits — the build did not

Building your project is free. Auto-Fix is not: it is a full AI run — the AI reads your files and the compiler output and writes corrected code — and it is billed like any other run. Pressing it three times on the same stubborn error is three runs.

Typing "fix the build error" in chat is the same run at the same price, only with less information: the button hands over the compiler output exactly as it came out, so it is usually the cheaper of the two. How Credits Work explains what makes one run cost more than another.

"Auto-fix blocked some writes"

Auto-Fix is only allowed to touch the files the build error actually points at. Before anything is saved, VyroAI compares the files the AI wrote against that list; a file outside it, written with no reason given, is thrown away rather than saved over your work. When that happens you get two messages — "Auto-fix blocked unrelated file writes", listing the files that were refused, and "Auto-fix blocked some writes".

The app does not send the fix again by itself. It offers one: a card appears in the chat panel proposing a second attempt, with the files that own the missing symbols added to what the AI is allowed to see — and then it waits. Nothing goes out and nothing is billed until you press Try a fix. Dismiss the card and the fix ends there, at no cost.

If you do accept, it goes out as "Retrying auto-fix with expanded context…" — and that retry is a second run, billed like one. One press of Auto-Fix followed by one press on the card therefore costs two runs. The card is only ever offered once per fix — a retry is never retried — so it cannot quietly run up a bill on its own.

If the same error keeps coming back, ask the AI to "investigate the root cause" instead — that switches from quick patches to a deeper architectural analysis.

Try it now: Describe your plugin and see the plan, free and with no account. Open the plugin generator

Where to next

5 of 7 in Power User · Last checked 11 September 2026 · Previous: The Model Benchmark