Agents that operate software built for humans, i.e., browsers, desktops, and applications, by combining LLM reasoning with visual understanding and simulated input. The agent acts as a proxy that performs actions the way a person would. - View it on GitHub