WebPilot is a flexible and autonomous multi-agent system that leverages the MCTS algorithm to simulate human cognitive abilities for strategic exploration and efficient web task execution.

[Paper]

# Python 3.10+
conda create -n WebPilot python=3.10; conda activate WebPilot
pip install -r requirements.txt
cd WebPilot
pip install -e .

We have intergrated the necessary WebArena environment code in the WebPilot repository. You can also install the WebArena environment separately.

cd webarena-main
pip install -r requirements.txt
pip install -e .

Start the docker according to the instructions in the WebArena repository.

Configure urls for the websites

export SHOPPING="<your_shopping_site_domain>:7770"
export SHOPPING_ADMIN="<your_e_commerce_cms_domain>:7780/admin"
export REDDIT="<your_reddit_domain>:9999"
export GITLAB="<your_gitlab_domain>:8023"
export MAP="<your_map_domain>:3000"

1. setting OpenAI API key

export OPENAI_API_KEY="<your_openai_api_key>"

2. Run the WebPilot pipeline

python run_WebPilot.py <task_id you want to run>

The WebPilot will now run the task to accomplish the task. Defaultly the brower will not be visible, but you can set it to be visible by setting the 'headless' parameter to False in the run_WebPilot.py file. A pdf showing the searching tree construction will show up to indicate the current progress. The results will be shown in the task_info folder. Each node in the pdf of searching tree refers to a screenshot of the webpage.

We show the successful trajectries of WebPilot on the WebArena. Coming soon.

If you find this work useful, please consider citing our paper:

@article{zhang2024webpilot,
  title={WebPilot: A Versatile and Autonomous Multi-Agent System for Web Task Execution with Strategic Exploration},
  author={Zhang, Yao and Ma, Zijian and Ma, Yunpu and Han, Zhen and Wu, Yu and Tresp, Volker},
  journal={arXiv preprint arXiv:2408.15978},
  year={2024}
}