UFO is a UI-focused dual agent framework for interacting with the Windows operating system. It executes user requests by understanding natural language and seamlessly navigating and operating within one or multiple applications. The framework consists of two agents: AppAgent and ActAgent. AppAgent is responsible for selecting applications based on user requests. ActAgent is responsible for iteratively executing operations within the selected application until the task is successfully completed. Both agents utilize the multi-modal capabilities of GPT-Vision to understand the application UI and fulfill user requests.