bmux
Agentic browsing
Give your agent a real browser you can see. Send commands to a specific pane, inspect the result, and monitor several workers in one workspace.
Video transcript
Agentic browsing.
I will send Page Down to scroll this pane.
Next, I will read the page DOM and filter it through a pipe.
Finally, I will click a link in this pane.
Start four independent profiles, then monitor their browsing and the commands sent to them.
Keep cookies, storage, and background browsing separate with profiles.
Control a pane from the shell
Every pane has an explicit ID. An agent can navigate, click, type, read the DOM, run JavaScript, and take screenshots. Commands return JSON and work in the background without taking your focus.
Send Page Down to scroll the target pane.
bmux key -t PANE_ID PageDownRead the page and filter its content through a pipe.
bmux dom -t PANE_ID | jq -r '.result.content' | grep 'Show HN:' | head -n 3Open a matching link in the same pane.
bmux click -t PANE_ID --selector 'a[href^="item?id="]'Watch several agents work
Put independent workers in split panes and follow their progress together. Commands target a pane rather than the currently selected page, so agents can work concurrently while you browse or inspect another pane.
Background browser controlSeparate logins and background work
Profiles keep cookies, storage, history, and connection settings separate. The bot profile keeps background pages running. Create another background profile for each independent account or worker; choose it when creating the pane.
Create a profile that keeps its background pages running.
bmux profile create research-agent --backgroundInstall on macOS
Get the latest Apple Silicon or Intel release with Homebrew
brew install --cask tonisives/tap/bmuxThe bundled CLI requires Node.js 22.12 or newer.
View release notes