This repository serves as a comprehensive, curated survey of the rapidly evolving field of Self-Evolving Agents. It targets researchers, engineers, and AI practitioners seeking a structured overview of the landscape, providing a centralized collection of surveys, papers, benchmarks, and open-source projects to accelerate understanding and development in this domain.

How It Works

The project categorizes self-evolving agents across three primary dimensions: Model-Centric Self-Evolution, focusing on improving the AI model itself through inference or training; Environment-Centric Self-Evolution, which enhances agent interaction with external knowledge and experience; and Model-Environment Co-Evolution, enabling simultaneous advancement of both the model and its operational environment. This structured taxonomy helps navigate the complex research space.

Quick Start & Requirements

This repository is a curated list of resources and does not have a direct installation or execution command. Users are directed to individual papers, benchmarks, and open-source projects for their specific setup requirements, which may include specific Python versions, deep learning frameworks (e.g., PyTorch, TensorFlow), hardware (e.g., GPUs), and large datasets.

Highlighted Details

• Features a detailed survey paper, "A Systematic Survey of Self-Evolving Agents: From Model-Centric to Environment-Driven Co-Evolution," providing a foundational reference.
• Organizes resources into key sub-areas including inference-based and training-based model evolution, static and dynamic knowledge/experience evolution, modular architectures, agentic topology, and model-environment co-evolution.
• Includes extensive lists of related survey papers, research papers with direct links, benchmarks, and open-source libraries relevant to self-evolving agents.
• Highlights emerging trends and applications in areas like automated scientific discovery, autonomous software engineering, and open-world simulation.

Maintenance & Community

The repository is actively maintained and welcomes community contributions. Users are encouraged to submit issues or pull requests for missing resources or new research findings. Contact is available via email at {xiangzhishang,yangchengyi}@stu.xmu.edu.cn and qinggangzhang@jlu.edu.cn.

Licensing & Compatibility

No specific open-source license is mentioned in the provided README content. Users should refer to individual linked papers and projects for their respective licensing terms and compatibility information.

Limitations & Caveats

As a curated list, this repository does not provide runnable code or a unified framework for self-evolving agents. Its value lies in its comprehensive indexing of existing research and projects, requiring users to explore and integrate individual components themselves.