AI assist

The assistant in the output pane - what each action does, exactly what it costs, which ones are free each day, and when you are not charged at all.

View as .md

The assistant, Revquix AI, reads the file you have open and answers about your code. It is in the output pane, next to the run results. There is no prompt box: every action is a button, and nothing is sent to a model until you press one.

The six actions and what they cost#

ActionCreditsFree per dayChanges your code
Explain error13Can suggest a patch
Explain code12No
Add comments1-Suggests a patch
Fix error2-Suggests a patch
Optimise2-Suggests a patch
Review2-No

Explain error and Fix error read your last run's error output, so they stay unavailable until a run has produced one. Each button shows its price before you press it, or "free" when your daily allowance or the cache covers it.

Large inputs cost more. An extra credit is added for roughly every 12,000 characters sent, counted after comments and whitespace are removed. That is around 400 to 500 lines of Java, so a typical file costs the base price.

When it is free#

  • Your daily free runs on Explain error and Explain code, as listed above.
  • A repeat of something you already asked. Same code, same action, same day - the answer is replayed from cache and costs nothing.

⚠ A cache replay does not consume one of your free daily runs either. Asking the same question twice is free the second time in both senses.

Patches#

Some actions propose a change rather than describing one. You get a diff - the current code beside the suggestion - and you choose.

Accepting applies the change to your file, where the editor's normal undo works on it.

What it can see#

The file you have open. The two error actions also get your last run's error output; the others do not. It does not know anything about your account beyond the code in front of it.

Getting better answers#

  • Open the file that matters before you press anything. The actions read the open file, so a small, focused file is cheaper and gets a sharper answer.
  • Use "Explain error" for a failure, not "Explain code". It is the one that reads the error.
  • Run it first. Without a run there is no error output, and the error actions have nothing to work from.

Was this page helpful?

Last checked against the product on . Behaviour changes are listed in the changelog.