The proposed agent, incorporating a "rewriting-resolution-review-revision" logical chain via three reasoning LLMs (functioning as the Consultant, Reviewer, and Programmer, respectively), is integrated in a collaborative and interactive manner.

For more details, see our arXiv preprint
- release streamlit version and employ langchain
src/
├── utils.py # General tool functions
├── FileHandler.py # File operation
├── code_executor.py # Code processing & execution
├── llm_agent.py # Client configuration
├── prompts_template.py # Prompt generation
└── re4_Agent.py # Workflow
If you have any questions on our work or implementation, feel free to reach out to chengao23@mails.ucas.ac.cn!
If you find this repository useful, please consider giving a star ⭐ and cite our paper.
@article{cheng2025re4,
title={Re4: Scientific Computing Agent with Rewriting, Resolution, Review and Revision},
author={Cheng, Ao and Zhang, Lei and He, Guowei},
journal={arXiv preprint arXiv:2508.20729},
year={2025}
}