2026-09-17

In-app browser: your agent works in the sites you are signed into

Sep 17, 2026

A browser now lives in the right panel: address bar, search, a row of bookmarks and history. Your sign-ins and open tabs survive a restart.

The browser belongs to the mark (thread or ticket). Ask again and the agent continues in the same tab.

Ask from the address bar hands the page you are looking at to your agent.

The agent reads the page, clicks, types, and presses Enter, Tab, Escape and the arrow keys.

You watch it work. A lime aurora lights the top of the page, the cursor moves, and one line tells you what it is doing right now.

Sending, submitting, paying and deleting ask first. Several items are picked and approved on one card.

Trust a site for an hour. Allow for 1 hour keeps approving on that thread and site for a while, then it asks again.

Touch the page and the agent pauses. Let it continue, or press Stop.

The agent never types your password. When a site needs a sign-in, it asks you and waits.

When it finishes, what it did and what it did not do stays in the thread.

1. Open the browser in the right panel of the desktop app.

2. Sign in to the sites you want it to use. Signing in is yours to do.

3. Press / in a thread, pick your agent and write the task. For example: read the three enquiries that came in yesterday and draft replies.

4. Watch it work. If something looks wrong, press Stop, or just use the page yourself and it pauses.

5. When the approval card appears, read it and allow or decline.

6. The result comes back to the thread, with the files or captures it made.

Sites have their own automation rules. It runs as the signed-in person, so use it the way each site allows.

The agent cannot use windows the operating system opens, like a sign-in window or a file picker. It asks you for those.