The Divi AI Agent is a chat built into the Divi 5 Visual Builder that carries out real work on your page - adding modules, changing settings, building whole sections - from a plain-language request instead of clicks through settings panels. This article is the first-run walkthrough: it assumes you already have an active Divi AI subscription, and it walks through opening the Agent, sending your first request, reading what it did, and undoing it if needed.
Before You Start
You'll need:
Divi 5. The Agent is built into the Divi 5 Visual Builder.
An active Divi AI subscription. The Agent is available to Divi AI subscribers only.
Note: Agent usage is billed differently from the rest of Divi AI. Instead of drawing on your Divi AI subscription itself, it draws on your AI Wallet balance, which your subscription funds with a monthly amount.
Opening the Agent
Open a page in the Divi 5 Visual Builder.
In the builder toolbar, click the AI Agent icon
.The chat window opens on top of your page, ready for your first message.
The Chat Window at a Glance
Before you type anything, it helps to know what each control in the composer (the message box at the bottom of the chat window) does. Point at any icon to see its name:
Commands
- reusable prompt snippets you've saved, so you can reuse a request without retyping it. Typing /in the composer opens the same picker.Rules
- standing instructions the Agent follows on every request, without you having to repeat them each time.Tools
- the catalog of actions the Agent is able to take on your site.Context
- reference material you attach to the conversation - a module, a preset, a page, and more - so the Agent knows exactly what you mean. Typing @in the composer opens the same picker.Attach a file
- add an image or a text file (JPEG, PNG, TXT, HTML, or CSV) for the Agent to look at.Change Mode - this button's own face shows your current mode, Ask or Build. Hover it to see the Change Mode tooltip, and click it to switch. Build is the default, and it's the mode this collection assumes unless an article says otherwise - the Agent can only make changes to your site while in Build mode.
Change Model - this button's face shows your current model. Hover it to see the Change Model tooltip, and click it to pick a different one.
Sending Your First Request
Try something concrete, like Add a three-column pricing section below the hero section, and send it.
While the Agent works, it posts a running log above your message - this collection calls it the Agent's Working notes. You'll see rows like:
Thinking - a short summary of what it's considering next.
Working - a generic status while it's using a tool or waiting on a step.
Tool used - a line naming the specific action it just took (for example, adding a row or a module), or a generic "Tool used" line when there's nothing more specific to show.
Todos - it's a running list of the steps it has planned and completed for your request.
You don't need to do anything while this runs. Watch your canvas instead - the builder updates live as each step completes, so you can see the pricing section take shape as the Agent builds it.
Reading What the Agent Did
Once the Agent finishes, its Working notes stay attached to that message so you can scroll back through them and see exactly what it did and in what order.
If the Agent split part of the work off to a sub-agent, you won't see that as a separate reply - its output folds into the same Working notes log as everything else, under the same generic Working status.
Undo an AI Agent's Change
Once the AI Agent finishes the task, if you are not happy with the result, you can revert the changes using the Visual Builder's Undo option.
Click on the Gear icon (located at the top-left corner).
Click the Undo option.
Notes: The Undo option reverts the page in the builder, but it doesn't save it for you. Saving is its own step, and the Agent always has to ask your permission before it can save, so nothing is permanent until you approve a save, whether that's the Agent's request or your own click.
Working Alongside the Agent
The chat window opens on the left side of the page's main canvas rather than blocking it, and it stays open while you work - you can click into the canvas and adjust settings by hand between requests, or while the Agent is still finishing a previous one, without closing the conversation.
Once you're comfortable opening the Agent and reading what it did, the next things to learn are Ask vs Build mode in more depth, and the prompt library of ready-made requests to try.