How Pluto works

Here is what happens between the moment you send a request and the moment your app updates. Understanding the loop helps you know what to expect and when to step in.

The build loop

When you send a request, PlutoAI works in a loop:

1. 1
   It reasons about the request

   The agent works out what you are asking for and what it needs to change, reading the relevant parts of your project first.

2. 2
   It makes the changes

   It writes and edits code across your project's files and runs any commands the work needs.

3. 3
   It checks its own work

   Before finishing, it asks: does the app build, does the live preview load, and is every planned step done. It will not call a job finished until those checks pass.

You can follow along as it works, and the live preview updates so you can see the result.

What you see while it works

A build is not a black box. As the agent works, you see its plan, the files it writes and edits, and the commands it runs, along with a short explanation of what it changed and why. That record is what lets you review a change instead of taking it on faith. See Working with your codebase.

When it hits an error

If the agent hits an error, it retries sensibly and works through the likely cause. Many errors clear on a second attempt. If it keeps hitting the same wall, it stops instead of looping forever, so you can step in and give direction, and so a stuck job does not drain your token allowance. See Troubleshooting.

Large builds

Large builds are not cut off partway. The agent works through all the planned steps before it finishes a run, so a bigger request still completes rather than stopping halfway.

You stay in control

At decision points the agent pauses and asks you before continuing, then resumes once you answer. See Agent approvals. It only uses a service once you have connected it, and it explains what it changed in the conversation, so you always know what happened.

Next steps