Timestamps: 00:00 - Intro 00:43 - Demo 02:00 - Overview 04:59 - UI-TARS Setup 11:26 - vLLM Setup 18:23 - First Run 22:22 - Multi-GPU Comments 23:57 - Closing Thoughts In this video, we explore UI-TARS, an open-source vision-language AI agent that can autonomously control your computer using natural language commands. Developed by ByteDance (TikTok’s parent company), UI-TARS leverages a powerful VLM backbone, allowing it to interact with your system through a graphical UI interface—all while running entirely offline on local hardware. We start with a live demo, showcasing UI-TARS in action as it executes real tasks based on simple prompts. Then, we break down the repository and technical overview, exploring how it works under the hood. From there, we walk through the full setup and installation process, including vLLM setup, to help you get this agent running on your own machine. Finally, we discuss multi-GPU considerations and real-world use cases for this intriguing AI agent.