How it works 

The AI agent works on the page you are editing. When you open it, it takes that page as its context. It reads the page and its content when needed, and every change it prepares applies to that page. Every request follows the same four steps. The AI agent never skips step 3 for a change.

AI agent window with the conversation list, the context line, a request, completed read steps, the Changes applied card with Undo, suggestions, and the message field highlighted

A completed request in the AI agent window.

The numbers in the screenshot mark:

  1. Conversations, filtered by This page or All pages.
  2. The context line: the page the AI agent works on, with its site, language, and workspace.
  3. Your request.
  4. Read steps the AI agent ran, each marked completed.
  5. The applied change with Open page, Edit, View on website, and Undo.
  6. What I can do: suggested next requests for this page.
  7. The message field and the AI provider used for the conversation. Choose the provider in the dropdown below the message field (for example Default: OpenAI) before you send the first message. See Change the AI provider.

The four steps of every request:

  1. Open on a page — Select the page you are editing, then click Ask AI Agent. The context line shows that page with its language and workspace. "This page" in your request always means the page shown there.
  2. Ask — Type your request and press Enter. The AI agent selects the tools it needs. Read tools (for example pages_get) run immediately and return an answer.
  3. Review and approve — If the request needs a change, the AI agent prepares a change card for the page, marked Preview — nothing changed yet. Click Apply to write it, or Cancel to discard it.
  4. Check the result — After Apply, the card shows Changes applied with links to the record and an Undo button.