Computer use capability lets agents operate a browser screen

The experimental computer_use capability lets an agent work a screen the way a person does. Its computer tool takes an action (click, type, key, scroll, drag, navigate or screenshot) and returns a screenshot, so any image-capable model can use it. The screen is the session’s Browserless browser.

On models with a native computer tool, the agent uses the provider’s tool instead; set native_tools to false to always use the function tool. max_actions_per_session caps actions per session, and navigation passes the egress guard. No call waits for a person by default: add soft_approval to record confirmations, or tool_approval to approve every call.

See Computer use. PRs #3922, #3968, #4064.