Computer-Use Agent
An AI agent that operates a browser or desktop directly — clicking, typing, and reading the screen — rather than calling a dedicated API, so it can complete tasks on sites and apps with no integration built for it.
Computer-use agents take a screenshot (or accessibility tree) as input and output actions like clicks, keystrokes, and scrolls, in a loop, letting them operate any website or desktop application the way a human would — including ones with no API at all. This makes them useful for automating long-tail workflows (internal tools, legacy software, sites that block scraping) that would otherwise need custom integration work, at the cost of being slower and less reliable than a direct API call when one exists. Most vendors in this category offer both a hosted, isolated browser environment (for safety and reproducibility) and a self-hosted/open-source option.
Last verified: