UI-TARS SOP

Appearance

Install ​

UI-TARS ships as three separate installable components depending on your use case.

Option 1: Python library (post-processing only) ​

Install the ui-tars PyPI package to parse VLM model outputs into executable pyautogui code. This does NOT download model weights; you supply the model inference separately.

Requires: Python 3.9+

bash
pip install ui-tars

Or with uv:

bash
uv pip install ui-tars

The CLI wraps a UI-TARS or cloud model backend and accepts natural-language tasks. Requires Node.js 22 or later.

One-shot (no global install):

bash
npx @agent-tars/cli@latest

Global install:

bash
npm install @agent-tars/cli@latest -g

Option 3: UI-TARS Desktop application ​

Download a prebuilt binary from the GitHub releases page (Windows, macOS, Linux):

https://github.com/bytedance/UI-TARS-desktop/releases

Latest release as of documentation: v0.3.0

Model weights (for self-hosted inference) ​

The VLM weights are hosted on HuggingFace. Pull the 7B model via the huggingface_hub CLI or git lfs:

bash
pip install huggingface_hub
huggingface-cli download ByteDance-Seed/UI-TARS-1.5-7B

Available checkpoints: UI-TARS-1.5-7B, UI-TARS-1.5-72B, UI-TARS-7B, UI-TARS-72B.

See the deployment guide for vLLM / HuggingFace Inference Endpoint setup.