Live browser
When a task needs a real website — searching listings, filling a form, checking your account — the assistant opens a browser and does it while you watch. Step in whenever a login, CAPTCHA or payment needs you.
Most questions are answered from a web search. Some jobs are not questions: "find 2-bedroom flats under 80k on Zameen", "message this seller on OLX", "check my order status on Daraz". Those need someone to actually open the site, click, type and scroll.
The live browser does that. It opens the site, works through it step by step, and posts what it found back into the chat. A panel beside the chat shows exactly what it is doing.
It works in chat, in agents and in Claw.
The live browser is part of the paid plans. On a plan without it, the assistant offers an upgrade instead of opening the browser.
When the assistant browses — and when it doesn't
You do not need to switch anything on. The assistant picks the right approach from how you ask:
| You ask for… | The assistant | Example |
|---|---|---|
| An answer — facts, news, prices, comparisons, reviews | Searches the web and writes the answer in chat | "Best phones under $800 right now?" |
| A task on a site — you name the site, want listings, or want something done | Opens the live browser | "Find Honda Civic 2018 listings on OLX Lahore" |
| To buy something | Finds products and shows shopping cards | "Order me a USB-C charger" |
Naming a site settles it: "on Zillow", "go to…", "check my … account" always opens the browser. Follow-up messages stay on the same site.
Before it starts, the assistant tells you in one sentence what it is about to do. The result arrives as a normal chat message when it finishes.
Live Browser mode
For bigger jobs, pick the Live Browser mode in the composer. The assistant may ask two or three quick questions first and shows you its plan as a checklist before it starts.
Where the browser runs
The panel header has three choices. Pick one before or during a task; your choice is remembered.
| Choice | Runs on | Your logins | Best for |
|---|---|---|---|
| Cloud (default) | A Volnyn server | No — it starts signed out | Public sites, research, listings. Keeps going with your laptop closed. |
| My Browser | Your own Chrome, through the Volnyn Chrome extension | Yes — your real logins and cookies | Anything behind your account: inbox, orders, social media |
| Computer | Your cloud computer | Yes — sign in once and it stays signed in | Recurring tasks on sites where you have an account |
Computer only appears when you have a cloud computer.
If a task clearly needs your own login and you are on Cloud, the assistant suggests switching. When it needs your Chrome, a card asks: "Authorize Volnyn to use a new tab from your Chrome to finish this on {site}" — click Connect, or No thanks to stay in the cloud. If the extension is not installed yet, the card offers Install instead.
The browser panel
The panel is titled Volnyn's computer and shows LIVE while the task runs.
- The screen — what the browser is showing right now, inside a browser frame with the tab name and address.
- Status line and step counter — what it is doing (for example 1 / 4 plan steps). Click the counter to see the step list.
- Timeline — scrub back through the run, or use Replay, Jump to start and Jump to live. Past runs can be replayed after they finish.
- Expand — widen the panel to full width, and Back to the chat to shrink it again. Drag the edge to resize.
For research-style runs, the panel also has Search and Editor tabs: the sites and steps it went through, and the answer as it is being written.
Stepping in
The assistant stops and asks you whenever it reaches something only you should do:
| It stops for | The box says | What to do |
|---|---|---|
| A sign-in page | Login needed | Sign in on the screen, then click Continue |
| A CAPTCHA or "press and hold" check | Solve the CAPTCHA | Complete it, then Continue |
| A payment or order confirmation | Confirm payment | Review and finish it yourself, then Continue |
| Something destructive or sensitive | Permission needed | Allow it, or Cancel |
Cookie banners and pop-ups it can close by itself do not interrupt you.
How you do the step depends on where it runs:
- Cloud — click on the screen in the panel. Your clicks are passed to the real page, which is enough for CAPTCHAs and consent walls.
- My Browser — the work happens in a real Chrome tab. Click Watch tab to bring it to the front, do the step there, and click Continue.
- Computer — the panel shows the computer's live screen. When it asks for help, click and type directly on it (paste with ⌘V / Ctrl+V, or the Paste button), then click Continue. Logins you make here are remembered for next time.
Take over (My Browser)
You can take the wheel at any time in My Browser mode:
- Click Take over. The agent pauses: "You are driving — click Continue when done."
- Click, type and scroll in the Chrome tab yourself.
- Click Continue. The agent picks up from whatever page you left it on.
Click Stop to end the task and close its tab.
Safety
- It never pays for you. At a checkout or payment step it stops and asks you to confirm and complete it.
- It does not claim success without proof. It only reports an order or booking as done when the page shows a confirmation.
- Destructive actions need permission. Deleting, cancelling or anything else hard to undo is asked first.
- Scheduled runs stop at sign-ins. An automation that meets a login wall reports that you need to sign in once; the schedule carries on afterwards.
Turning web browsing off
Web browsing is on by default. To switch it off:
- Open Connectors from the chat composer.
- Find Web Browser — "AI browses the live web for you — research, prices, news, any site".
- Click Disconnect. You will see "Web Browser disconnected — the AI will not browse for you."
With browsing off, a task that needs a site shows a card: "Web browsing is switched off for your account, so I cannot open any site." Click Turn it on and the assistant picks the task straight back up, or Leave it off.
Click Connect on the same row to switch it back on.
Limits
- Each run has a step limit and a time limit. When it runs out of time it wraps up with what it found so far.
- The cloud browser is shared. If it is busy you will see "The cloud browser is busy with another run. Try again in a minute, or switch to My Browser."
- If a run goes quiet for too long, the panel shows "The run stopped responding. Press Reconnect, or send the task again."
Tips
- Name the site when you know it. It saves the assistant a search.
- Be specific about filters — city, price range, dates, size. It applies them on the site like you would.
- Use My Browser or Computer for anything personal. The cloud browser has no access to your accounts.
- Ask for the output you want. "…and give me a table with price, link and seller".
FAQ
Can I watch what it is doing?
Yes. The panel shows every step live, and you can replay a finished run from its timeline.
Does it see my passwords?
In My Browser and Computer modes you sign in yourself, on the real page. The assistant does not ask for or store your password.
Why did it answer from search instead of opening the browser?
Questions are answered faster from search. Name a site or ask for a task on one — "on Amazon", "check my…" — and it will browse.
Why can't I see the Computer option?
It only appears when you have a cloud computer.
Does browsing cost credits?
Browsing is part of the task it belongs to. The chat message, agent run or Claw task is charged as usual.
Next steps
- Install the Chrome extension so tasks can use your own logins.
- Set up a cloud computer for tasks that should run while you are away.
- Give an agent the Web Search and Web Fetch abilities and a schedule.