PrompteryxBeta
    ⌘K
    PlatformSupport
    Menu
    IntroductionInstall Desktop & extensionQuick StartPlatform Overview
    OverviewWhat you can makeAI Browser Agent in the chatModels and AutoConnectorsSchedulesProjectsExplore and publishingShare linksMemory and SkillsPersonal agent (beta)Welcome, settings and skinsPlans, credits and top-upsPrompteryx DesktopGroups (pilot)Phone app (early access)
    OverviewCreating WorkflowsTesting & DebuggingScheduling
    OverviewGetting StartedCommandsRecordingSettingsStructured OutputParallel AgentsTips & Best Practices
    OverviewInstallationSidebar AgentUsing Attached FilesExecute on Current TabExporting to Visual Studio
    Overview
    OverviewOpen PageClickClick CoordinatesFillExtract TextScreenshotWaitPress KeyFocus ElementBlur ElementSelect TextHighlight ElementSet ViewportGet Bounding BoxScrollHoverSelect OptionAssertionsCookiesEvaluate JavaScriptExample Workflows
    OverviewPrompteryx Browser AgentOpenAI
    OverviewSchedule TriggerWebhook Trigger
    OverviewGoogle SheetsGoogle CalendarConnect Hub (3,000+ apps)Example Workflows
    AI Vision SettingsChrome Automation ProfilesCloud BrowserBrowser Profiles & Persistent ContextVideo RecordingExecution LogsSharing & CollaborationEdit History
    Selectors GuideVariables
    Start hereGet an API keyRun the AI browser agentDrive a cloud browserRun workflows from anywhereUse connected apps (Connect Hub)SDK (TypeScript)CLIMCP for AI assistants
    REST APIWebhooksMCP toolsCloud Browser APICloud Browser quickstartMigrate from Browserbasen8nZapierMake.com

    Need help? Contact Support

    1. Documentation
    3. App
    5. Browser Agent

    AI Browser Agent in the chat

    Ask in the main chat and an agent opens a real browser, clicks, types and reads for you, and tells you how it went.

    Choose AI Browser Agent under the message box, or just ask for something that needs a website ("go to example.com and tell me the main heading") and the chat starts the agent for you. A bar appears above the message box with the agent's settings.

    The app home with AI Browser Agent chosen: the heading asks Where should the agent browse?, and a bar above the message box shows the profile, the model, AI Vision, Record and More settings
    With AI Browser Agent chosen, the bar above the message box sets where it runs, the model, AI Vision and recording.

    Where it runs

    Cloud Browser

    A fresh browser on our servers. Nothing to install, and it keeps going if you close your laptop. The profile menu shows how many cloud minutes you have left.

    Use a cloud profile to stay signed in to sites between runs.

    Your own Chrome

    Pick one of your Chrome Profiles and the agent works in your own Chrome, with your logins and cookies, using free local minutes.

    This needs Prompteryx Desktop and Prompteryx in Chrome. If they aren't set up, the Chrome profiles show Not connected and a Set up button that walks you through three steps: install Prompteryx Desktop, sign in to it, and add Prompteryx in Chrome. Use Cloud Browser instead skips it.

    The settings bar

    Profile
    Where it runs: Cloud Browser, a cloud profile, or one of your Chrome profiles.
    Model
    Which agent does the work (below).
    AI Vision
    How sharp the screenshots the agent sees are. Sharper reads small text better and costs a little more.
    Record
    Records the run as a video you can watch or download afterwards. Some plans show Upgrade here.
    More settings (the sliders icon)
    Settings for this message only: structured JSON output, thinking level, final step quality, safety checkpoints, a residential proxy and its country (Cloud Browser), the most credits this run may use, and parallel agents (first message, Cloud Browser).
    ⚙ (the gear)
    Your saved AI Browser Agent settings, used for every run.

    While a browser session is open these are locked to it. They apply to the next session; press End Chat to change them now. The thinking level lists the levels the chosen model really has, with its own default marked (default). Every setting is explained in AI Browser Agent settings.

    Models

    Prompteryx Core
    Cheaper and faster. The default.
    Prompteryx Prime
    The best all-rounder.
    Prompteryx Swift
    The fastest and cheapest.
    Prompteryx Prime+
    The newest brain.
    Prompteryx Jev
    Ultra-fast and text-driven. Beta.
    Gemini, Claude and GPT
    Google's, Anthropic's and OpenAI's own computer-use models, for when you want a specific provider. The premium ones cost more per step.

    Every model in this list works in your own Chrome as well as in the Cloud Browser: the Prompteryx models, Gemini, Claude and GPT. If a model can only run in the Cloud Browser, its Chrome profiles are greyed out with Cloud Browser only for this model. To change the model used when you don't pick one, open Settings → Model defaults and change the AI Browser Agent row.

    Watching it work

    The live view shows the browser as the agent uses it (Cloud Browser Live View, or Live Browser View for your own Chrome). Press Hide live view to give the chat the whole width, and Show live view to bring it back. On a phone it opens as a sheet.

    • Stop: press Stop in the message box (or Esc). The agent stops after its current step.
    • Safety checkpoints: before something that can't be undone, such as submitting a form or paying, the agent asks with a Continue / Stop card.
    • Keep chatting: send another message and the agent carries on in the same browser, signed in where it was.
    • And then…: ask for the next step in the same message, such as "find the three cheapest flights, then put them in a spreadsheet". When the browser part ends with an answer, the chat carries on with the rest by itself.
    • End Chat (or End Chat and Cloud Browser) closes the browser and stops the clock on cloud minutes.

    Runs carry on in the background

    Leaving a chat or starting a new task doesn't stop a run. Chats that are still working show a spinner and running · step N in the sidebar, and you can go back to them at any time. In Prompteryx Desktop you get a notification when a long reply finishes.

    In your own Chrome, Prompteryx Desktop runs the whole task itself, so it keeps going even after you close or refresh the tab. You can open the chat on any device to follow it, and press Stop from anywhere. This is Keep local runs going in the background in Settings → AI Browser Agent, which is on by default; switch it off to run the task inside the tab instead (closing the tab then stops it). It needs Prompteryx Desktop to be running and signed in, and Claude and GPT need Prompteryx Desktop 1.5.6 or later for this. Prompteryx Jev always runs inside the tab.

    The outcome card

    Every run ends with one card that tells you plainly how it went.

    Answer
    It finished. The card says what it did, step by step, and gives the answer.
    Partly done
    It got part of the way. You see what it found so far and why it stopped.
    Needs you
    It stopped because something needs a person, such as a sign-in or a payment.
    Didn't finish
    It could not do the task, for example it ran out of steps or the session reached its time limit. The card says why.

    When a run didn't fully succeed, the card offers what to do next: Try again, up to three Try with … buttons that send the same task to a different model (a stronger one first, then another provider), and Open in Visual Studio to turn what the agent did into a workflow you can edit and run again. Every card also has See how to re-use this as a repeatable / recurring automation. If the result is unclear, a quick AI check decides which card to show; it uses a small number of your credits.

    An AI Browser Agent run in the chat: the steps on the left ending in an outcome card that says All done, lists what it did and gives the answer, with the Cloud Browser live view on the right
    A finished run: the outcome card (All done, What I did, Answer) next to the live view.

    Doing it again

    • On a schedule: press Schedule in the message box before you send. See Schedules.
    • As a workflow: Refine in Visual Studio at the top of the run, or Open in Visual Studio on the outcome card.
    • Share it: Share task to Explore lets other people run the same task. See Explore.

    Cost

    The agent's model is paid with AI credits for each step; Cloud Browser time uses your cloud minutes, and your own Chrome uses free local minutes. Set Max credits for this run in More settings to cap a single run. The full-page agent at /AI-browser-agent still works and has the same settings; see AI Browser Agent overview.