How do I start the browser on my own computer?

Install the app from samreshuuu.ru/download, open it, switch the environment next to the message box from the cloud to your computer, and write "open … in the browser on my computer". The first time, confirm in the "Agent browser" dialog.

Will the agent see my passwords and open tabs?

The browser runs in a separate profile: your normal tabs, sessions and passwords saved in the browser are left alone. The agent neither sees nor types passwords: when a site asks you to sign in, it hands control to you.

Does the browser only work in the app?

The agent opens cloud pages without the app. To work on your computer, the app must be installed and open there. You can select a connected computer and use the chat from the web version too.

Why does nothing open on my computer?

Check the environment picker next to the message box: the intended computer must be selected. If it is unavailable, open Devices & sessions in settings: the status and explanation under its name give the next step. A lost connection alone does not mean the app is closed.

Working on your own computer

The agent opens cloud pages without the app. That is enough to look something up, read a site or pull data off public pages.

The app adds a second browser — the one on your computer. It opens on your screen, works from your IP and keeps your own logins to websites. That is what you need where the cloud cannot reach: a supplier portal, a government service, an internal admin panel, a site with no API, a page behind the corporate network.

Getting started

Install the app

Open samreshuuu.ru/download and install the build for your system — macOS or Windows. If your system warns that the developer is unverified on first launch, the same page walks you through opening it in a couple of clicks.

Check that the chat runs on this computer

Next to the message box there is an environment picker: "In the cloud" or the name of your computer. Pick the intended computer. If it is unavailable, open Devices & sessions in settings and follow the step under its name, such as enabling the connection or signing in again. No link means the app is already open but its connection to the service is not getting through; the cause, when known, is shown there too.

Say what needs doing in the browser

In plain words: "open the supplier portal and export the stock levels", "sign into the account and file the request", "check in the browser on my computer what happened to the order". There is no "turn on the browser" button — the agent launches it when the task calls for it.

Allow the launch

The first time, an "Agent browser" dialog appears — click Allow. You are asked once for the whole app.

A browser window then opens on your screen, and a live stream titled "Local browser · live" appears in the chat — you see every click even with the window minimised.

What the agent does in the browser

opens pages and follows links;
clicks, fills and submits forms;
pulls numbers, tables and statuses off pages;
uploads files to a site and downloads exports;
takes screenshots of what it sees.

The result comes back as usual — an answer in the chat, plus a file if you asked for a table or a report.

Checking local pages

Select the computer that holds the page and ask the agent to open it. On Windows and macOS, the app's agent browser opens HTML files from the working folder the agent has access to, and local servers such as http://localhost:5173 or http://127.0.0.1:8080. These addresses need no separate permission in browser settings.

If the page needs HTTP for modules, data requests or paths such as /assets/, the agent can start the project's dev server and check its local address. Opening an HTML file directly follows the browser's rules for local files.

For another computer or service on your local network, such as 192.168.1.20:8080, the agent offers an access card. Approve it in the app on the computer running the browser. You can also manage these permissions under Settings → Capabilities → Agent browser → Local addresses (host:port). The browser's own control ports remain unavailable.

The agent first checks whether this profile is already signed in. When a site asks you to sign in, the agent hands control to you: you enter the login, password, verification code, CAPTCHA or passkey yourself, and the agent neither sees nor types them. Sign in in the same browser window the agent is using, then click "I'm done, continue" on the card or above the screen. The agent checks the current page before continuing. A sign-in page alone does not switch to "You're driving": that mode follows a handoff or your intervention.

The session is kept in the agent's profile until the site requires you to sign in again.

Your normal browser is left alone

The agent works in a separate profile: your tabs, bookmarks, sessions and saved passwords stay untouched, and the system keychain is out of its reach. Everything it does happens on your screen, and you can step in at any moment.

Worth knowing

The browser on your computer needs the app running there. You can select a connected computer from the web version too. Cloud mode opens pages in the cloud with its own profile and sign-ins; to use your computer's IP and agent profile, select it next to the message box.
Chromium-based browsers only. Google Chrome, Microsoft Edge, Brave, Vivaldi, Opera, Yandex Browser, Chromium and Chromium-Gost. For sites that require GOST encryption (the Federal Tax Service cabinet for companies, the EIS procurement cabinet), ask the agent to open Chromium-Gost: CryptoPro CSP and the CryptoPro plug-in must be installed on the computer, and you enter the token PIN yourself. The app uses your saved choice, then a supported system default, then another installed browser from this list. You can ask the agent to open a particular installed browser from the list. Safari and Firefox are not supported yet.
The computer has to be on and the app open. It lives in the Dock or the tray and keeps working in the background; you can enable "Launch at login" in the settings. If the app is closed, the chat offers to switch back to the cloud.
Scheduled tasks run in the cloud. Your computer may be off at the scheduled moment, so recurring tasks do not rely on the local browser.
Marketplaces. For data and actions available to your account through an API, the agent uses integrations. If the API does not cover the action or it is unavailable to your account, you can ask it to use the local browser. Completion depends on account access and the site's checks; a CAPTCHA requires a handoff to you. Connecting a browser alone does not confirm sign-in or guarantee that a reply can be published.
Banks. The agent works with statements and payments through the bank connection in integrations. Bank agreements usually prohibit sharing the login and password with third parties, so you sign in to online banking yourself.
Closing the browser window is fine. The agent will ask to relaunch it on its next step. Quitting the app itself also closes the agent's window.
The agent does not obey website content. Text on a page is data, not instruction: it never follows instructions found on a third-party site or shares secrets at their request.

If something goes wrong

The computer is unavailable: check its status in Devices & sessions in settings. A signed-out app, a disabled connection, a closed app and a broken link need different actions; the explanation under the computer's name gives the next step. Your selected computer stays selected when its connection drops. To work in the cloud, choose it explicitly; local files are not transferred to the cloud by that choice.
"This action needs a newer version of the desktop app": update the app on the computer this chat uses. Repeating the same action before updating will not help; other available actions may still work.
"The agent could not take control of the browser": the browser opened but did not let the agent drive it in time. Most often a dialog in its window is waiting for an answer: answer it and repeat the request.
"The browser closed while starting": the browser process the app started quit during launch. Repeat the request; if it keeps happening, check that this browser opens normally on its own.
"The agent's port is taken": the port the agent drives the browser through is held by a program the app did not start. Retrying will not help until you close that program or restart the app.
"The browser cannot be automated": the requested browser is unsupported, such as Safari or Firefox, or no installed Chromium browser was found. Ask for a supported browser; if none is installed, install one from the list above.