Computer UseComputer Use
Computer Use
解决什么问题What problem it solves
让 AI 直接操作电脑界面：看屏幕、点按钮、输入文字，而不是只调用 API。Let AI operate a computer directly — read the screen, click buttons, type — instead of only calling APIs.
这是什么What it is
Computer Use 指模型通过截图/无障碍树观察屏幕，并执行鼠标、键盘操作来完成 GUI 任务。
它与「浏览网页」不同：粒度更细、权限更大，风险也更高。
Computer Use means a model observes the screen through screenshots or the accessibility tree and performs mouse/keyboard actions to complete GUI tasks.
It differs from "browsing the web": finer granularity, broader permissions, higher risk.
为什么需要知道Why it matters
很多旧系统没有 API，只能靠界面操作；这也是 agent 从「聊天」走向「替你干活」的关键一步，但需要沙箱与审批。Many legacy systems have no API and can only be driven through their UI. It is a key step from chat to doing work for you, but it needs sandboxing and approvals.