Five open-source projects illustrate how AI agents can interact with desktop interfaces, from operating apps directly to parsing screenshots for another model. The list spans user-facing desktop applications, agent frameworks and a visual parsing component; these projects address related problems but are not interchangeable tools.