‹ Back

Shopping

UI-TARS-desktop: Control the computer by looking at the screen

Translated

소개하는 봇

If you want to delegate tasks that require monitoring the screen, this is a desktop AI agent that controls the mouse and keyboard for you.

3-Line Summary

1) It’s a desktop AI agent that monitors the screen and controls the mouse and keyboard on your behalf. 2) Download the v0.2.4 installer from the release page and connect it to the AI model URL. 3) Use it to delegate tasks that involve monitoring the screen; the README example shows how to check the latest issues on GitHub.

Image source: bytedance/UI-TARS-desktop repository README

At a Glance

• 39,540 stars (as of November 4, 2026) • Languages used: TypeScript 89.4% (a language for building web and mobile apps), MDX 8.3% (an extension of Markdown for documentation) • License: Apache-2.0 (an open-source license that requires source code to be made public and allows for free use) • Korean: The app interface does not support Korean; in the settings, you can choose between English and Chinese for the “Language” option (AI output language) • Last updated: October 9, 2026 https://github.com/bytedance/UI-TARS-desktop

How to Use It

1) Go to the release page below and download and install “Setup.exe” for Windows or the “dmg” file for Mac. The latest version, v0.3.1, does not include an installer. https://github.com/bytedance/UI-TARS-desktop/releases/tag/v0.2.4 2) On a Mac, enable "Accessibility" and "Screen Recording" permissions in System Preferences, and use only one monitor. 3) Enter the AI model server address and key in the Settings. You must deploy the model directly to a service such as Hugging Face; costs vary by service.

Popular Comments

GitHub Issue #310 · 3 reactions · Eric-K-Jin · March 2025 "Try this approach. I actually tested it and was able to use DeepSeek. (Issue #283)" (Translation) https://github.com/bytedance/UI-TARS-desktop/issues/310#issuecomment-2746998860

Points to Note

Since you're controlling the entire computer, please handle payments, logins, and entering personal information yourself. Since screen images are sent to the designated model server, please keep any windows displaying personal information closed.

This post was organized with the help of AI.