The browser
Player One has a real web browser of its own. Not a search box that fetches pages as text, but a full Chromium that your agents open, click around in, fill forms in, download from, and sign in to, exactly as you would. Most of the time it works out of sight. Whenever you want, you can watch it, and take the keyboard.
What it does for you
Say "get last month's statement from the courier portal", "fill in the customs form on the shipping site from this order", or "check the price of this part across the three suppliers".
The agent opens the site, finds its way to the right page and does the work: reads what is there, fills what needs filling, presses what needs pressing, and brings back the result. Anything it downloads lands beside the chat for you to save. It also takes screenshots, captures a page as a PDF, lifts a table off a page, and records a short video of a site behaving badly, so "show me what the booking page does when I pick Tuesday" gets a clip rather than a description.
How it finds what to press
Say "add the grey jacket to the basket and tell me the total".
The agent does not have to read the whole page and work out what to press. It names the element in words, such as "the Add to basket button under the jacket", and Player One picks the matching one on the page and presses it. A login or a form runs as one short sequence of up to eight steps, so "sign in with these details" is one move rather than a look and a click for every field. Where Enter is what sends or adds something, as in a search box or a to-do list, a typing step can press it.
Afterwards the agent can say what the page should show, such as "the basket shows one item". Player One looks at the page again, waits a moment if it is still loading, and answers yes, no or not sure. To read one fact, such as a total or a status, the agent asks the page a question and gets back the passage that answers it, copied from the page. If nothing on the page answers, it says so and takes a full look instead.
None of this puts the whole page into the conversation, so a job usually runs faster and costs less. When Player One is not sure which element was meant, it presses nothing and lists the closest few, and the agent chooses. A wrong press shows in the result, so the agent checks it and puts it right. What an agent may do without asking is unchanged: Stay in control still applies.
Who does the choosing
A small, fast model makes these choices, not the model your agent is thinking with. If you have a judgement provider connected, it does the job quickly. If you do not, a fast model from the providers you have connected does the same work, a little slower. Admin > Agent Models > Advanced Routing (/admin/models/routing) has a Browser locate rule if you would rather choose the model yourself. Models and providers covers both.
That model is sent the page. It sees the buttons and fields when it picks what to press, and the page's text when it checks a result or answers a question, including pages you are signed in to. When you connect a judgement provider on Admin > Providers (/admin/providers), the screen says so, and you confirm that the provider handles your data under its own terms. An agent running in a sandbox never sends its pages to a judgement provider.
Watching it work
Say "show me what you are doing".
The same browser appears in the preview pane beside the chat, live, as a view-only window. You see the pages change as the agent works. A Take control button hands you the mouse and keyboard when you want them, and Hand back returns them. Nothing needs installing: the window is streamed into your own browser by Player One itself, over the same secure connection as the rest of the app.
Signing in without giving it your password
When a site wants a human, the agent gets as far as the sign-in page and then puts that browser in front of you. You click inside it, type your password, deal with the code from your phone, and press Hand back. The agent carries on in the same browser, on the page you left it on. While you hold the keyboard the agent can look but not touch, so "does this look right?" is a safe question. Your password went into the website in that window and never passed through the agent or the chat.
Log in for it once, let it work from then on is the worked example.
Your own browser instead
If you have Sidekick's browser extension connected and you are chatting from that browser, your agent can work in the tab you are looking at: your own browser, with your own sign-ins. Nothing is handed over and nothing expires, so for a site you already use it is the easier route, and your agent offers it first.
Your browser shows its own banner while Player One is in a tab, and Stop in the panel ends it at once. If the page changes under a sequence of steps, the agent looks again and tries that step once more before it stops. Sidekick has the controls. The browser described on this page is for work that carries on without you, and for when no extension is connected.
It remembers logins
Each login is kept in a named browser profile, so the next time that portal comes up it is already signed in. One profile is one identity: if you have two accounts on the same site, ask for a separate profile for the second and the agent sets it up. A project can have its own default profile, so the agents on the accounts project use the accounts logins and nobody else does. A profile remembers that you are signed in, never what you typed. Passwords belong in the vault.
Where to see it
Admin > Resources (/admin/resources) has a Browser section showing each profile, whether it is open, how many tabs it has and how much memory it is using. Each extra profile is a separate browser and costs a few hundred megabytes while it runs, which is why the agent mentions the cost before creating one, and why browsers left idle are closed for you. The login stays: a closed browser reopens signed in.
Good to know
- A passkey or password manager on your own computer does not follow into the window. Use the site's other route: a code by message, an authenticator app, an emailed link or a security key.
- On a phone the live window works, but the floating Hand back button is desktop only. Say "done" in the chat instead.
- If you close the window instead of pressing Hand back, the browser is still running. Say "done" in the chat.