hai-drivers

Web, desktop and shell drivers for H Company's computer-use agents.

pip install "hai-drivers[web]"          # browser driver (Selenium)
pip install "hai-drivers[playwright]"   # browser driver (Playwright, async)
pip install "hai-drivers[desktop]"      # desktop driver
pip install "hai-drivers[desktop-cli]"  # `desk`: desktop commands for an agent's shell
pip install "hai-drivers[browser-cli]"  # `web`: browser commands for an agent's shell
pip install "hai-drivers[cua-cli]"      # `cua`: desktop commands for an agent's shell, through cua-driver

desk click X Y, desk key ctrl+s, desk screenshot; desk aliases puts one executable per command next to desk so the bare click, key, screenshot work as well.

web goto URL, web text, web click @e4, web screenshot: one web executable drives the Chrome already running with --remote-debugging-port=9222, through a warm daemon holding the Playwright driver.

cua windows, cua state, cua click 12, cua type "text", cua shot: one cua executable reads each window's accessibility tree and acts on numbered controls or screen pixels through a warm cua-driver daemon, with a receipt per action saying what the driver could verify. cua skill prints the manual.

LocalCodeSandbox runs an agent's shell and file calls in a folder on this machine, with the permissions of the user running it.

This package is consumed by the hai-agents SDK to let agents control a local browser, desktop or shell; it is not meant to be used on its own.