---
title: browser-use-annotator
canonical_url: "https://www.modelscope.cn/studios/kongquyu/browser-use-annotator"
md_url: "https://www.modelscope.cn/studios/kongquyu/browser-use-annotator.md"
repository: kongquyu/browser-use-annotator
chinese_name: "浏览器操作标注工具"
last_updated: 2025-06-10
sdk_type: gradio
sdk_version: 5.29.0
downloads: 0
stars: 3
---

# browser-use-annotator

> browser-use-annotator - kongquyu 在 ModelScope 创建的在线 Demo。A web-based tool for annotating browser use trajectory for VLMs

kongquyu/browser-use-annotator 是 ModelScope 魔搭社区上的在线可交互 Demo（创空间），基于 gradio 5.29.0 构建，中文名为「浏览器操作标注工具」。

- **Repository**: kongquyu/browser-use-annotator
- **SDK**: gradio
- **SDK version**: 5.29.0
- **Downloads**: 0
- **Stars**: 3
- **Last updated**: 2025-06-10

Source: https://www.modelscope.cn/studios/kongquyu/browser-use-annotator

---

# Browser Use Annotator

A web-based tool for annotating browser interactions to create high-quality training datasets for vision-language models like **Qwen2.5-VL** and **UI-TARS**.

![Browser Action Annotation Tool Demo](img/demo_1.png)


## Perfect for Training

This tool is specifically designed to create training data for:
- **[UI-TARS](https://github.com/bytedance/UI-TARS)** - ByteDance's GUI interaction agent
- **Qwen2.5-VL** - Alibaba's vision-language model
- **Other multimodal agents** requiring web interaction understanding

## Installation

1. Install the package:
   ```bash
   uv sync
   ```

2. Install Chrome browser for Playwright:
   ```bash
   playwright install chrome
   ```

3. Start the annotation tool:
   ```bash
   # Add your startup command here
   uv run app.py
   ```

## Usage

1. **Start a New Task**: Enter your task description in the "Current Task" field
2. **Load Target Website**: Submit the URL you want to annotate
3. **Begin Annotation**: Interact with the website normally - all actions are automatically captured
4. **Add Manual Annotations**: Use the action buttons (WAIT, Scroll Up/Down) for specific behaviors
5. **Review Actions**: Check the conversation log to verify captured interactions
6. **Save Data**: Save your annotated session for model training
7. **Reasoning Traces Annotation**: navigate to Reasoning Annotation page at top-right, load the dataset and annotate the reasoning for each action.

## TODOs

- [ ] Connect to external hosted browser services (e.g., [AgentBay](https://agentbay.console.aliyun.com/))
- [ ] Export data in UI-TARS training format

## Acknowledgments

This project builds upon and includes code from: https://github.com/trycua/cua