UI-TARS is an open-source stack designed to automate GUI interactions on local and remote computers using vision-language models. It features a desktop application and CLI that interpret natural language instructions to perform complex tasks through visual recognition and MCP tool integration.