UI-TARS Desktop is part of the TARS multimodal agent stack, which focuses on agents that can understand visual interfaces and operate computers and browsers. The desktop application provides a native GUI agent built around the UI-TARS model, while the related Agent TARS project extends the same ideas into a broader multimodal agent environment with CLI and web interfaces.

The stack combines vision, GUI interaction, browser control, shell tools, and MCP integrations so agents can complete workflows that resemble human computer use. Agent TARS supports headful and headless execution, event-stream-based debugging, hybrid browser control using visual and DOM information, and the ability to connect additional real-world tools through MCP servers. Remote computer and browser operators extend these workflows beyond the local machine.

UI-TARS and Agent TARS are aimed at developers exploring computer-use agents, browser automation, multimodal reasoning, and general task execution. They are useful for workflows such as navigating websites, collecting information, operating graphical applications, and combining visual interaction with conventional agent tools when an API-only approach is insufficient.