Use the Agent for Hard Edits

Overview

Click-to-edit works when an element's value is written plainly in the source. A lot of real code isn't like that. A heading might come from t('hero.title'), a list of cards might be built with .map(), and an image might be passed in as a prop. There's no single line for a click to change. For those, Studio hands the edit to the agent chat, which reads the code and makes the change where it actually lives.

How it works

When an edit routes to chat

When you click an element whose value is indirect, Studio doesn't fail silently. It opens the chat with a message already filled in, pointing at the element you selected, so you can describe the change and send it. You'll see this when you click:

• A translated string, like {t('key')}, that lives in an i18n catalog.
• An item rendered inside a .map() over a list or data source.
• A value passed in as a prop or referenced by a variable, like src={logo}.

The editor panel also tells you when it has sent an edit to the agent, so you know the click was received and routed, not dropped.

📘 Media pending

Screenshot of a routed edit in the chat hasn't been captured yet.

Asking for an edit directly

You don't have to click first. Type any request into the chat and send it. If you have an element selected on the canvas, Studio adds that context to your message, so "make this bigger and bold" applies to the thing you picked. The agent works in your repo, so it can touch the files it needs to, not just one line.

📘 Media pending

Screenshot of a typed chat request hasn't been captured yet.

Choosing the agent

The chat runs on Claude. The agent switcher at the top lets you pick between your own Claude and Blu, Intempt's agent, once you're signed in. On the base tier you use your own Claude, so the chat works before you sign in.

📘 Media pending

Screenshot of the agent switcher hasn't been captured yet.

Watching the work

As the agent runs, the chat shows each step. Tool runs appear inline with a running, success, or error state, so you can see what the agent is doing in the code rather than waiting for a black box. A cost meter under the composer tracks the tokens used this session, and the dollar cost when there is one. You can stop a run in progress from the composer.

📘 Media pending

Screenshot of the chat during a run hasn't been captured yet.

Fixing a broken state with chat

The chat isn't only for content edits. When the dev server fails to start, or a publish build fails, Studio gives you a Fix with chat button that hands the captured error to the agent to diagnose and fix. It's the same chat, pointed at the problem.

📘 Good to know

After the agent finishes an edit, Studio can offer a suggestion for a useful next step, including edit-time SEO and accessibility hygiene like missing alt text or a missing meta description. Applying a suggestion drops it into the chat composer as your next request.

Where to go next

• Edit a page visually for the clicks that don't need the agent.
• Preview and publish your site to review the agent's changes and ship them.
• Example: edit your pricing page for a worked edit.
• Website Studio for the overview.
• Desktop App Overview for the rest of the desktop app.