UI-TARS-desktop
Multimodal GUI agent stack that operates your desktop and browser
What it is
ByteDance's TARS project bundles two agents: UI-TARS Desktop, a native app that uses the UI-TARS vision-language model to control a local or remote computer and browser from natural-language instructions, and Agent TARS, a CLI and web UI agent that combines GUI control, vision, and MCP tools. People who want an AI to click through websites, fill forms, or run multi-step desktop tasks can run it with their own model endpoints. Developers can also build their own GUI automation agents on the included SDK.
