A browser tab shows a web page beside the session, such as your dev server. The agents can use the browser too: they open pages, click, type, and take screenshots to test their work. You point them to the parts of a page to change with annotations.

A task or a workspace session can have at most 8 browser tabs. The popup windows of a page count too.

Open a web pageCopy Anchor Link

1. At the end of the panel tabs, select +, then Browser.
2. Type an address, for example localhost:5173, and press Enter.

A browser tab opens only http and https addresses. If you type no scheme, OpenDucktor adds http:// for localhost and private network addresses, and https:// for all other addresses. The address bar does not search the web.

To open a link of the transcript or a terminal in a browser tab, select the link, then OpenDucktor browser. Shift+Cmd+click opens it in a new tab directly. On Windows and Linux, use Ctrl.

Let the agents use the browserCopy Anchor Link

The agents of the task use the browser with the OpenDucktor browser tools. Each action shows as a Browser card in the transcript. On the page, a ring shows where the agent clicks or types, and a dot pulses on the icon of the tab.

An agent sees only two kinds of tabs:

• Its own tabs. An agent can open a tab by itself. The tab opens at the end of the right panel with the Agent badge. You can use it and close it.
• The tabs that you share. To let the agents use one of your tabs, turn on Share with agents in its toolbar. The agents cannot see or use your other tabs.

In a tab that it can use, an agent can read the page, click, type, select, scroll, take screenshots, read the console, and answer the dialogs of the page. It cannot upload files, and it cannot close a tab that you share. If you stop sharing a tab, the actions that the agent sent but did not do yet fail.

An agent never sees the permission requests or the certificate warnings of a page. When a page asks for one, you answer it. See Site permissions.

In the desktop app, all the tabs of a workspace use the same cookies. Sign in to your app once in your own tab, and the agent tabs are signed in too.

The agents of the sessions that OpenDucktor starts get the browser tools: Spec, Planner, Builder, QA, and workspace sessions, on OpenCode, Codex, and Claude Code. An external MCP client does not get them.

Annotate a pageCopy Anchor Link

An annotation is a note on one or more elements of the page. It tells the agent what to change, and the agent gets the elements and a screenshot with your next message.

1. Select Annotate.
2. Select an element of the page. To add or remove more elements, Shift+click them.
3. In Tell the agent what to change, type your note, then select Save.
4. Send a message to the agent. Your pending annotations go with it.

To check your notes before you send them, select Pending annotations. A command such as /compact does not send them.

OpenDucktor keeps pending annotations only in memory. A reload of the app deletes them.

Site permissionsCopy Anchor Link

When a site asks for a permission, such as the camera or your location, a bar shows above the page. Your choice applies to this site in all the tabs of the workspace, also the agent tabs. OpenDucktor keeps it until you remove the workspace.

To change a choice, select Site settings at the left of the address. It lists the saved choices for Camera, Microphone, Location, Notifications, and Clipboard read. Select Reset, and the site asks again.

When a page does not showCopy Anchor Link

In the browser runnerCopy Anchor Link

Everything above is for the desktop app. The browser runner works in another way, because the page shows in your own browser:

• Your tabs show only on this computer. A tab shows the page in a frame of the browser runner. This works only when you open the browser runner at localhost, 127.0.0.1, or [::1]. At another address, for example from another computer, the tab shows This app cannot show pages.
• The agents cannot use your tabs. Your browser shows the page, so the agents cannot reach it. Share with agents, annotations, and site permissions are not available. Your browser shows the permission requests and the developer tools of the page.
• Agent tabs are hidden. The agents still use the browser: they open their pages in a Chrome that runs on the host, without a window. You see their actions only as Browser cards in the transcript. Their first use downloads this Chrome, about 200 MB.
• Agent tabs do not use your logins. Their Chrome has its own cookies, so an agent is not signed in to the sites of your tabs.

A client of a remote host and the mobile app show no web pages.

NextCopy Anchor Link