Appearance
UI-TARS
Category: OSS Agent (GUI Automation / Vision-Language Model)
UI-TARS is an open-source multimodal agent built on a vision-language model (VLM) that can perceive and interact with graphical user interfaces across desktop, mobile, and gaming environments. It was developed by ByteDance and is the underlying model powering the Agent TARS ecosystem.
The model interprets screenshots, reasons about the UI state, and outputs executable actions (clicks, typing, scrolling) as structured pyautogui code. It achieves state-of-the-art results on GUI interaction benchmarks including OSWorld, ScreenSpot, and AndroidWorld.
Key facts
| Field | Value |
|---|---|
| Maintainer | ByteDance / ByteDance-Seed |
| License | Apache 2.0 (free, open-source) |
| Stars | 11,000+ |
| Category | OSS Agent - GUI automation via native VLM |
| Pricing | Free |
| Model weights | HuggingFace: ByteDance-Seed/UI-TARS-1.5-7B |
Variants
| Package | What it is |
|---|---|
ui-tars (PyPI) | Python post-processing library: parse model outputs into pyautogui code |
UI-TARS-desktop | Electron desktop app with local/remote operators |
@agent-tars/cli (npm) | CLI wrapper that routes tasks to a UI-TARS or cloud model backend |
Official links
- GitHub: https://github.com/bytedance/UI-TARS
- Desktop app: https://github.com/bytedance/UI-TARS-desktop
- Research paper: https://arxiv.org/abs/2501.12326
- Website: https://seed-tars.com/
- Releases: https://github.com/bytedance/UI-TARS-desktop/releases
Safety note
Because UI-TARS can autonomously navigate GUIs including CAPTCHA flows, ByteDance includes a safety notice: the model must not be used for unauthorized access, credential theft, or any activity that bypasses access controls. Internal safety evaluations are ongoing. Always operate within the terms of the target system.