The Grok Bot browser scores 8.1 out of 10 on my scorecard: the best sign-in handling I've seen documented for an agent browser, held back by a shared computer and a few steps that still need you.
It's the browser on the Bot's own cloud computer, not a browser on your laptop.
Every score below comes from what xAI's docs and changelog say, with a one-line reason.
These are editorial ratings, not lab benchmarks.
If you want my team to tell you which agent tools fit your business, book a free SEO strategy session.
The verdict in one line
Use it for logins, portals, PDFs and downloads where no app exists, and keep a connector for everything that has one.
The scorecard
| Feature | Score /10 | Why |
|---|---|---|
| Sign-in security | 9.5 | Vault, 1Password and Secure Form values never reach the Bot (Vault docs) |
| Approval cards | 8.5 | Plain-word cards since v0.70.2, with Allow once, Always allow and Deny |
| Human handoffs | 8 | Clear Take over and I'm done flow for SSO, passkeys and CAPTCHAs |
| Session persistence | 8.5 | Logins survive browser restarts and closed windows |
| v0.78.1 typing | 7.5 | One character at a time with mouse moves; slower but steadier |
| Connector offers | 8 | Offers the app on eight named sign-in pages |
| Outputs: PDF, downloads, screenshots | 8 | Page-to-PDF, original file names, full-size tall screenshots |
| Isolation between Bots | 6.5 | One shared computer; screens aren't security boundaries |
The average across the eight rows is 8.1.
Feature by feature: why each score landed where it did
Sign-in security: 9.5
This is the strongest part of the product.
A Bot signs in with a login saved in Vault first, then a connected password manager such as 1Password, then a Secure Form.
The approvals and security page says those values are filled into the page without being shown to the model.
A filled Secure Form even says so: "Secret values were never shown to your Bot."
1Password goes further, since v0.77.0: you approve each login inside 1Password and it fills the Bot's browser directly.
One-time codes sent by email get picked up from your connected Gmail or Outlook, or the Bot's own inbox.
I take half a point off because 1Password won't fill cards and only offers logins from your Personal or Private vault.
Approval cards: 8.5
Auto Review is on by default and checks each action.
Browser cards now describe the action in plain words, like "Open a new tab in Grok Bot's browser", per the v0.70.2 entry on the Grok Bot changelog.
Always allow saves a rule, and that saved rule now applies the same way on every tool.
Deny stops the Bot cold until you send another message, a change in v0.75.5.
Your own Ask first rules always win over Allow rules.
The score isn't higher because broad rules are easy to write, and the docs themselves warn against "allow everything in the browser".
Human handoffs: 8
The Bot hands you the computer for single sign-on, passkeys, a stubborn CAPTCHA, approvals that happen on a second device, QR codes, bank card verification and human-only sites.
Press Take over, clear the one blocked step, press I'm done, then tell the Bot to continue.
Since v0.75.2 you get 20 minutes to finish a sign-in, a Copy link button, and Reopen link to pick the same attempt back up.
It loses points because laptop passkeys, Face ID and Windows Hello don't work on the Bot's computer, per troubleshooting.
Hardware keys work on macOS and Windows, but not yet on Linux.
🔥 Want the setup that makes these scores work in practice? Inside the AI Profit Boardroom you get step-by-step video trainings, the installable Agent OS, four coaching calls a week and 3,400+ members rating new agent tools as they ship. → Join the AI Profit Boardroom
Session persistence: 8.5
Sign in once and it sticks.
v0.62.0 keeps the Bot signed in when its browser restarts.
The 0.75.2 improvements keep site logins after the last window closes and fix Apple ID sign-in forms.
v0.52.0 stopped the browser offering to save passwords after it signs in, which keeps credentials where they belong.
The catch is that persistence is shared: signing in for one Bot signs in all of them.
v0.78.1 typing: 7.5
The October 27 release made the browser enter text a single character at a time, with mouse movement, on each website it fills.
The changelog doesn't give a reason, so I won't invent one.
What it means in practice: form fills look like a person typing, and they take longer.
The same release fixed approved actions failing on pages that hadn't changed, and the computer view now follows the tab in front.
Good changes, but slower fills keep this at 7.5.
Connector offers: 8
Land on a sign-in page for Google Ads, Meta Ads, TikTok for Business, YouTube, Cloudflare, Coinbase, Gusto or Railway, and the Bot now offers the matching connector.
During handoffs for Google, GitHub or QuickBooks, it offers the connector too, and connecting ends the handoff.
That's the right call, because the computer and apps docs say apps are more reliable than clicking through a site.
Outputs: PDF, downloads and screenshots: 8
v0.70.2 lets a Bot save a web page as a PDF from its browser.
The same release keeps downloads under their own file names.
0.75.2 keeps screenshots full size, including tall pages.
The files and results page adds that a Bot can push a downloaded file to cloud storage or onto an email draft.
I'd score it higher if the docs said more about download folders, but they don't.
Isolation between Bots: 6.5
This is the weak spot.
Every Bot on your account shares one cloud computer: the same cookies, sessions, files and command-line credentials.
Each Bot gets its own screen, and the docs say the screens aren't security boundaries.
Deleting a Bot doesn't remove browser sessions either.
Isolation between users is strict, with dedicated, hardware-isolated computers, but inside your own account you have to manage it yourself.
📞 Weighing up which agent tools are worth it for your site? Book a free SEO strategy session and my team will review your pages, your rankings and the work you'd hand to an agent. → Book your free session
Reliability fixes that lift the overall score
The changelog shows steady browser work across releases.
- 0.64.0+: crashed tabs recover, large uploads work, and stuck pages time out at 25 seconds.
- v0.57.0: a Bot's browser stays connected when other Bots on the same computer use theirs, and waits longer for slow pages.
- v0.75.2: the Bot warns you before a browser task that will take a few minutes.
- 0.70.2+: scripts the Bot runs in its browser no longer fail over extra or out-of-range settings.
None of these earn their own row, but together they explain why the persistence and handoff scores sit above 8.
How it behaves on pages it didn't write
A browser agent reads a lot of text it can't trust.
xAI's answer is simple: text from somewhere else, such as a web page, doesn't count as your request.
Before an action it can't undo, like sending, posting, buying or deleting, the Bot checks your own messages for approval, a v0.70.2 change.
When a site blocks automation, demands a fresh login or shows a CAPTCHA, the FAQ says the Bot should hand the step to you rather than get around it.
That's the behaviour you want, and it supports the 8.5 on approvals.
Approvals when you aren't watching
Timing matters for unattended work.
A card from a task you started in chat waits for you and doesn't expire.
A card raised by a routine or other background work expires after 30 minutes, and the Bot doesn't take the action.
Sending a new message closes any open card in that chat, and so does changing your Auto Review settings.
An expired card still lets you press Always allow next time if it offered a rule.
These rules are clear and predictable, which is most of what you want from an approval system.
Teach a task: a bonus, not a score
When Teach a task is available, you record one browser workflow from the computer view and the Bot drafts a skill from it.
The skills docs cap the recording at ten minutes and say the rollout may be gradual.
I left it off the scorecard because not everyone has it yet.
Limits you should know before you buy in
Some sites block traffic from cloud computers.
The fix is Route traffic through this computer, which sends the Bot's browser through your desktop's connection while the app is open.
Team admins should know an app block leaves the service's website reachable; only Network Controls, an Enterprise feature, shuts it, per the security docs.
A Vault card pays at checkout without asking you each time.
And Grok Bot itself needs the Premium, Super or Ultra plan on The X Subscription, per the plan page; Plus doesn't include it.
How it compares with the wider computer-use feature
The browser is one slice of the Bot's cloud screen.
For the desktop, recovery and shared-screen side, read our Grok Bot computer use review.
This scorecard rates the browser alone.
Who should use it
- Owners and operators who sign in to the same portals every week.
- Agencies pulling ad and analytics screens, especially now the Bot offers Google Ads, Meta Ads and TikTok for Business connectors.
- Researchers who want pages saved as PDFs and full-size screenshots in a folder.
- Anyone who wants an agent to sign in without ever handling the password.
Who should skip it
- Teams that need hard separation between agents on one account.
- Workflows that live entirely inside apps with good connectors.
- Jobs that need a passkey or Face ID every time, since those stay on your own devices.
🔥 Get the verdicts before you spend a week testing Inside the AI Profit Boardroom you get the trainings, the installable Agent OS, four weekly coaching calls and 3,400+ members comparing agents on real jobs. → Join the AI Profit Boardroom
Final verdict: 8.1 out of 10
It's the most careful agent browser in the docs I've read, thanks to Vault, 1Password, Secure Forms and plain-word approval cards.
Shared sessions are the trade-off, so set narrow Ask first rules and sign out of what you don't need.
If you want a hand deciding where it fits, book a free call with my team.
Also On Our Network
- 🌐 The seven-step setup I run before the Bot's browser touches a login
- 🌐 Where the Bot's browser saves a small business owner real hours
- 🌐 Five copy-paste browser prompts and a weekly web workflow for your Agent OS
- 🌐 Which models to pair with the Bot's browser, with real bench scores
- 🌐 The complete Grok Bot Browser guide on the AI Profit Boardroom blog
FAQ: Grok Bot Browser Verdict
What is the best feature of the Grok Bot browser?
Sign-in security. Logins from Vault, 1Password or a Secure Form are filled into the page without the Bot seeing them, which is why it scores highest on this scorecard.
What is the biggest weakness?
The shared computer. Every Bot on your account uses the same browser sessions, cookies and files, and the docs say separate Bots are not a security boundary.
Is the Grok Bot browser better than a connector?
No, and xAI agrees. The docs say a connected app is more reliable than clicking through a website, and since v0.78.1 the Bot offers the connector on several sign-in pages.
How were these ratings made?
They are editorial scores based on what xAI's Grok Bot docs and changelog say each feature does, with a one-line reason for every score. They are not benchmark results.
📺 Video notes + links to the tools 👉
🎥 Learn how I make these videos 👉
🆓 Get a FREE AI Course + Community + 1,000 AI Agents 👉
About Julian
I'm Julian Goldie, an SEO entrepreneur, author and founder of the AI Profit Boardroom, which has 3,400+ members.
I help business owners scale with AI agents, automation and SEO.
- I've built a 7-figure SEO and link building agency, Goldie Agency, with a team of 70+ SEO experts.
- I've grown a YouTube channel to 400,000+ subscribers.
- I wrote "Link Building Mastery".
- My Udemy courses have taught over 29,000 students.
→ Get my best AI training inside the AI Profit Boardroom
Set three Ask first rules before your first run, then join the AI Profit Boardroom to see how members use the Grok Bot browser.
