Computer and browser
Each Concierge job gets its own private computer and browser. How it works, how you sign in to a site for the agent, and what it can and can't read.
Last updated: 2026-10-19
When a job needs real work, such as crunching a year of orders, building a spreadsheet or reading a page on a supplier's site, the agent gets a computer of its own. It's a clean, private machine that exists only for that one job and is shut down when the job ends. Nothing runs on your laptop, and the agent never uses your own browser.
A private computer per job
- Started when needed. The agent turns on a computer only when the job calls for it. In the conversation you'll see "Getting a workspace ready", and "Opening a browser" if it also needs a browser.
- Your data, read-only. The computer gets a read-only copy of the store data the job needs. The agent can analyze it with Python, spreadsheets and charts, but it can't change your store from there.
- No internet from the command line. Commands the agent runs on the computer have no network access. Web pages are reached only through its browser, and only the sites allowed for the job.
- Shut down afterward. When the agent replies, the job is stopped or a newer request replaces it, the computer is shut down and its copy of your data is deleted.
- What carries over. Only files the agent saves as outputs. Everything else on the computer, including any site sign-ins, is gone after the job.
What the browser can do
The agent's browser can open a page, read its text and links, find buttons and fields, click, type, pick options and scroll. Some things are deliberately held back:
- Only allowed sites. Each job has an exact list of secure (HTTPS) sites it may visit, including the sites you sign in on. Redirects or requests to anything else are blocked and reported.
- It reads and looks; it doesn't submit. Opening pages and clicking can only read. Typing fills a field without sending its form, and anything that would write to the site is blocked.
- Downloads and uploads ask first. A download needs your approval. Where an upload is supported, the agent prepares it, shows you the exact account, folder and file name, and sends it only after you approve. An upload is sent at most once; if the result is unclear, the agent checks instead of sending it again.
- It never types secrets. The agent can't type into password fields, and never types a password, one-time code or card number.
Sign in for the agent
When a page needs you to sign in, the agent hands its browser to you. This works in Admin chat, for the teammate who made the request.
The agent asks you to sign in
It pauses and posts: "Open the browser and sign in there. Then reply here when you are ready to return browser control. Do not send your password or verification code in chat." the browser is a link.
Open the link
A page called Sign in to your browser opens. Click Connect. When it says "Connected", click Refresh view to see the agent's browser.
Sign in
Click into the browser view and sign in with your keyboard and mouse, including any verification code. The status line reads "You have browser control."
Reply in the conversation
Go back to the conversation and reply, for example "Done, I'm signed in." That returns control to the agent and closes the sign-in page. The agent checks the page to confirm you're signed in, then carries on.
Your password and codes are typed only on the sign-in page. They don't go into the conversation, the agent's steps or its memory, and the page doesn't log them. The sign-in lasts only for that job; the next job asks again.
Only the teammate who made the request can open the link, and only while the agent is waiting for that reply. Once you reply, the job ends or someone stops it, the link stops working.
What it can and can't read
It can read
- The read-only copy of your store data prepared for the job
- Files you attach, files from playbooks and files saved from earlier jobs that it brings back in
- Pages on the sites allowed for the job, including pages you signed in to for it
- Public web pages and web search results, through separate web tools that never send customer personal details in a search
It can't read
- Your own browser, its tabs, saved passwords or cookies
- Sites outside the job's allowed list
- Other jobs' computers, or anything from a previous job that wasn't saved as an output
- Passwords typed on the sign-in page
Text on web pages is treated as information, never as instructions. A page that says "ignore your rules" doesn't change what the agent does.
See what it did
Under an agent's reply in Overview, click Show work. Work behind this answer lists the Data read and the Commands run on the computer. Progress lines in the conversation, such as "Reading the page" or "Uploading the file", show browser steps as they happen.
Troubleshooting and FAQ
"This browser handover is no longer available to you."
The link only works for the teammate who asked, while the agent is waiting. If you replied already or the job ended, ask the agent again and use the new link.
"Browser disconnected" or "Connection lost"
Click Reconnect. After "Connection lost", look at the page before you retype, because your last keystroke may have reached it.
Why does it ask me to sign in every time?
Each job's computer is new, and sign-ins are deleted with it. That keeps your accounts from staying open on a machine nobody is watching.
The agent says a site was blocked
The page tried to reach a site outside the job's allowed list. Tell the agent which site it needs, and it can start again with that site allowed.
Can it fill in and send a form on a website for me?
It can fill fields so you can see them, but it doesn't send forms. Use the capabilities on the agent's page for actions in Chargezen and connected apps.
Related
Guides
On the product page
Learn more
Related Articles
Need more help? Contact support