**RepoAgent: An LLM-Powered Open-Source Framework for Repository-level Code Documentation Generation** **Qinyu Luo, Yining Ye, Shihao Liang, Zhong Zhang, Yujia Qin, Yaxi Lu, Yesai Wu, Xin Cong, Yankai Lin, Yingli Zhang, Xiaoyin Che, Zhiyuan Liu, Maosong Sun** **Abstract:** Generative models have shown significant potential in software engineering, particularly in tasks like code generation and debugging. However, their application in code documentation generation remains underexplored. To address this gap, we introduce RepoAgent, an open-source framework powered by large language models (LLMs) designed to proactively generate, maintain, and update repository-level code documentation. Through both qualitative and quantitative evaluations, we validate the effectiveness of RepoAgent, demonstrating its ability to produce high-quality documentation. The code and results are publicly accessible at <https://github.com/OpenBMB/RepoAgent>. **Introduction:** Developers spend a significant portion of their time on program comprehension, and high-quality code documentation plays a crucial role in reducing this time. However, maintaining documentation is also time-consuming and resource-intensive. Early attempts at automatic documentation generation focused on providing descriptive summaries for source code, but they had limitations in summarization, guidance, and passive update. RepoAgent addresses these issues by leveraging LLMs to generate comprehensive, practical, and up-to-date documentation for entire repositories. **RepoAgent:** RepoAgent consists of three key stages: global structure analysis, documentation generation, and documentation update. It uses a project tree to maintain all code objects and their semantic hierarchical relationships, and it leverages reference relationships to enhance understanding. The documentation generation stage employs a sophisticated strategy to generate fine-grained documentation, while the documentation update stage integrates with Git to track and update documentation automatically. **Experiments:** We conducted experiments on 9 Python repositories of varying scales, using different LLMs as backends. Human evaluation and quantitative analysis showed that RepoAgent outperformed existing methods in identifying reference relationships, format alignment, and parameter identification. The results indicate that RepoAgent can effectively generate and maintain high-quality documentation, enhancing productivity and collaboration in software development. **Conclusion and Discussion:** RepoAgent is a promising tool for automating the generation and maintenance of code documentation, improving team collaboration and software quality. Future work will focus on expanding its applicability to multiple programming languages, reducing the need for human oversight, and addressing security and privacy concerns.**RepoAgent: An LLM-Powered Open-Source Framework for Repository-level Code Documentation Generation** **Qinyu Luo, Yining Ye, Shihao Liang, Zhong Zhang, Yujia Qin, Yaxi Lu, Yesai Wu, Xin Cong, Yankai Lin, Yingli Zhang, Xiaoyin Che, Zhiyuan Liu, Maosong Sun** **Abstract:** Generative models have shown significant potential in software engineering, particularly in tasks like code generation and debugging. However, their application in code documentation generation remains underexplored. To address this gap, we introduce RepoAgent, an open-source framework powered by large language models (LLMs) designed to proactively generate, maintain, and update repository-level code documentation. Through both qualitative and quantitative evaluations, we validate the effectiveness of RepoAgent, demonstrating its ability to produce high-quality documentation. The code and results are publicly accessible at <https://github.com/OpenBMB/RepoAgent>. **Introduction:** Developers spend a significant portion of their time on program comprehension, and high-quality code documentation plays a crucial role in reducing this time. However, maintaining documentation is also time-consuming and resource-intensive. Early attempts at automatic documentation generation focused on providing descriptive summaries for source code, but they had limitations in summarization, guidance, and passive update. RepoAgent addresses these issues by leveraging LLMs to generate comprehensive, practical, and up-to-date documentation for entire repositories. **RepoAgent:** RepoAgent consists of three key stages: global structure analysis, documentation generation, and documentation update. It uses a project tree to maintain all code objects and their semantic hierarchical relationships, and it leverages reference relationships to enhance understanding. The documentation generation stage employs a sophisticated strategy to generate fine-grained documentation, while the documentation update stage integrates with Git to track and update documentation automatically. **Experiments:** We conducted experiments on 9 Python repositories of varying scales, using different LLMs as backends. Human evaluation and quantitative analysis showed that RepoAgent outperformed existing methods in identifying reference relationships, format alignment, and parameter identification. The results indicate that RepoAgent can effectively generate and maintain high-quality documentation, enhancing productivity and collaboration in software development. **Conclusion and Discussion:** RepoAgent is a promising tool for automating the generation and maintenance of code documentation, improving team collaboration and software quality. Future work will focus on expanding its applicability to multiple programming languages, reducing the need for human oversight, and addressing security and privacy concerns.