More AI browser: import your data, research websites and test your app

A working browser beside your conversation #

More AI 0.1.50 includes the built-in desktop browser and browser-data import. The exact controls and import support depend on your app version and operating system. Use it for website research in Home/Cowork or for a running app in Code. In Code, open Browser from the side-panel menu; when an agent opens a page, its browser activity card also opens the corresponding conversation’s panel.

The agent can navigate, read page content, find elements, follow links and enter text, subject to the task’s permissions. You can inspect the same page, pause agent control and take over. Tabs and site storage belong to that conversation, so another task starts with a separate profile unless you have chosen reusable imported defaults.

Start with an observable result: “Open these three product pages and compare their export formats. Keep the source URLs and flag anything you cannot verify.” Reading the page is part of the agent’s work; you do not have to attach a screenshot for every page it reads.

Import data from a browser you already use #

Open Browser data, the database icon in the browser toolbar. You can choose a detected browser profile, select a profile folder, or open a compatible export file. The built-in import does not require the Chrome extension.

Import is a manual action in the app. The agent does not get an import command or a tool for reading your password vault. Starting an import pauses its browser control; the browser remains paused until you resume it.

1. Choose the source profile or export. If the source is locked or unavailable, close that browser and retry, or use its supported export format.
2. Review the categories: bookmarks, history, cookies and saved passwords. Passwords are normally unselected; a password-only export selects its sole category.
3. Review “Use in new conversations”. It is on by default; turn it off to keep this import for the current conversation. Existing conversations retain their separate site data.
4. Start the import and read the per-category imported and skipped counts. A partial result can be useful; the report explains unsupported encryption, expired cookies or other skipped records.
5. Open the needed site, sign in again if required, then resume agent control. You can cancel an import; closing and reopening the dialog preserves its progress.

What can transfer from Chrome, Firefox and Safari #

Import support follows the actual profile format and encryption, not just the browser’s name. Chrome, Edge, Brave, Vivaldi, Opera and many other Chromium-family profiles are discoverable, alongside Firefox-family browsers and Safari. A detected profile does not guarantee that every category can be imported.

Compatible files include password CSV, bookmark HTML, cookie TXT/JSON, history CSV/JSON and Safari export ZIP. A browser that exports passwords and bookmarks may not offer a cookie or history export. The dialog shows which categories are available for the selected source.

• Chromium on macOS: history and bookmarks can import directly; supported cookie and password formats may require native Keychain authorization.
• Chromium on Windows: history and bookmarks can import directly. Supported DPAPI and older encrypted entries may transfer; app-bound v20 entries are skipped. Use password CSV or sign in inside More AI when needed.
• Chromium on Linux: secret import depends on the source format and a usable desktop keyring. Missing helpers or unsupported wallets require an export or a fresh sign-in.
• Firefox family: history, bookmarks and supported cookies can import directly; saved passwords use an exported CSV. Container and partitioned cookies are skipped.
• Safari: history and bookmarks can import directly when accessible; passwords use CSV or a compatible Safari ZIP. Cookies need a compatible export.
• Extensions, payment cards, passkeys, OTP secrets, downloads and another browser’s localStorage or IndexedDB are not transferred. Imported cookies do not guarantee that a site keeps you signed in.

Useful tasks beyond browsing #

A good browser task names the source, the output and how to check it. The examples below are workflows to try with your own accessible pages; they are not measured performance claims. Add files or a working folder when the result should be a document or spreadsheet.

• Marketing research: “Compare these five landing pages: audience, promise, evidence and call to action. Produce a table with one source URL per row and separate observations from suggestions.”
• Accounting and operations: “Read this visible dashboard and my exported working copy. Reconcile the date range, currency and totals, and list discrepancies for review.” Verify the original records before using the result for reporting.
• Product research: “Read the official documentation for these three tools. Compare the required features, record the review date and mark missing evidence instead of guessing.”
• Form preparation: “Help fill this form using the details I provide, then show the completed fields for my review.” State explicitly whether any final submission is authorized.
• Local website QA: “Check navigation, this form and the mobile layout on localhost. Report reproduction steps and verify each proposed fix in the running app.”

• Web search and source checking — Find sources first, then inspect the pages that answer the question.
• The file library — Organize the files and deliverables used in a task.

Preview and check a local app #

In Code, the browser can use the project’s supported dev-server configuration, such as .claude/launch.json or a package.json dev script. Start a detected server from the panel, or open its localhost URL when it is already running. The activity controls let you inspect and stop the associated work.

Ask the agent to check the changed flow, not merely open the home page. It can read the rendered interface, use supported console/network diagnostics and resize the viewport for mobile, tablet or desktop checks. Screenshots require an image-capable model; a resized desktop browser is not a real-device test.

• Code agent: from repository to verified change

Capture and annotate a visual problem #

Use the camera button to capture the page, mark the relevant area in the annotation editor and add it to the conversation. Include the viewport, the action that caused the problem and the expected result. “The label overlaps this button at this width” is more useful than “make it better”.

The image becomes task context for the selected model. Choose a model that can receive images when visual inspection matters. The agent can also take its own supported screenshots during browser work; your annotated capture is a way to point out a particular detail.

Point at the exact element #

Use the element picker in the toolbar and select the control or text you mean. More AI adds the selected element as context for your message, so you can request a precise change without copying a selector by hand.

For example: “Keep this button’s action, but fix its alignment with the input.” Ask the agent to reread the page after navigation or a layout change and verify the result. An element reference helps locate the problem; it does not prove the change works.

Saved data, permissions and local versus cloud work #

In Browser data → Saved data, search history and bookmarks, open an entry, choose a saved account for login filling, or remove records. The scope selector distinguishes this conversation from defaults for future conversations. A saved login is filled only on a matching site and does not submit the form; multiple accounts require a choice.

Reusable defaults seed new profiles once. They do not continually overwrite existing conversations, restore cookies you deliberately cleared or make session cookies permanent. Closing a panel or tab keeps that conversation’s site data; deleting the conversation removes its browser profile. Future-conversation defaults are managed separately.

Sites receive ordinary web requests. Page content and images used by the agent may go to the selected model provider. Enabled synchronization, cloud workflows, AI observability and replay have their own data handling; review Privacy settings for sensitive work. A separate local browser profile is not a promise that every task stays on your computer.

This guide describes the desktop browser. Its interactive agent tool is not the browser for an unattended or remote run. Cloud browser sessions and cloud execution are separate account features with their own access and limits. The desktop app is free; BYOK model usage is billed by your provider. Available cloud features are temporarily free during beta.

Frequently asked questions #

Can I import passwords and cookies from my current browser?

Yes, for supported profiles or export formats. Choose the categories in Browser data and inspect the import report. Support varies by operating system, browser version and encryption; Windows app-bound entries and some cookie types are skipped.

Does importing cookies mean I never need to sign in again?

No. Expired, partitioned or device-bound sessions may not transfer. Complete a fresh login in the browser when a site requests it.

Can the agent use the browser without the Chrome extension?

Yes. The embedded desktop browser has its own agent tool and profile. The Chrome extension connects your external browser and is a separate option.

Will the desktop browser keep working when my laptop is closed?

Its work depends on the desktop app and computer remaining available. Cloud execution and cloud browser sessions are separate features; importing a local profile does not move it to the cloud.