Review every change before it lands

An agent that edits your game is only useful if your game still feels like yours afterward. Kiwari is built around a simple rule: you see what it wants to do before it does it, and you decide how much it can do on its own.

Every edit is a diff you can read

When Kiwari changes a script, you get the change as a diff right in the chat, with color, expandable in place. Added lines, removed lines, nothing hidden behind "done". You approve it and it lands in Studio. You reject it and your place never saw it.

This has a side effect people do not expect: reading the diffs teaches you. You watch how a DataStore save is structured or how a remote is wired, in your own game, in the style of your own code.

You set the autonomy

Different sessions want different amounts of supervision:

• Approve each edit when the place matters or the change is delicate. Nothing touches your game without you seeing it first.
• Run everything when you are prototyping and speed matters more than review. You can still read every change afterward in the chat.
• Ask mode when you only want answers about your project, with no edits at all.
• Plan mode when the work is big. Kiwari writes a plan first, you review and adjust it, and only then does it build.

Settings, Permissions decides the default, and you can tighten or loosen it whenever. Script edits can be undone, and Studio's own undo still works for everything else.

Why this shape

Plenty of tools can generate Luau. The hard part of working with an agent is trust: a fast edit you did not see is a bug you will hunt next week. Keeping every change visible and reversible is what makes it comfortable to hand Kiwari real work on a real game, not just a throwaway baseplate.

Pair it with playtesting and the loop closes: Kiwari proposes, you review, it verifies the change in Play, and it tells you plainly what it checked and what it could not.

Related: How Kiwari playtests your game · Projects and memory · Download