Awesome-Self-Evolving-Agents: A Curated List for AI Agent Research

Introduction

The Awesome-Self-Evolving-Agents repository, maintained by XMUDeepLIT, is a meticulously curated collection of resources dedicated to the burgeoning field of self-evolving agents. This repository stems from a systematic survey paper, "A Systematic Survey of Self-Evolving Agents: From Model-Centric to Environment-Driven Co-Evolution," and aims to provide a continuously updated hub for researchers and practitioners. It covers a wide array of materials, including foundational survey papers, cutting-edge research articles, performance benchmarks, and practical open-source projects.

Self-evolving agents represent a transformative paradigm in AI, enabling systems to autonomously enhance their capabilities across three primary dimensions:

Model-Centric Self-Evolution: Focuses on improving the agent's internal model through inference-based and training-based evolution.
Environment-Centric Self-Evolution: Enhances agent interaction with external knowledge and experience, encompassing static knowledge, dynamic experience, modular architecture, and agentic topology evolution.
Model-Environment Co-Evolution: Explores the simultaneous evolution of both the agent's model and its environment, including multi-agent policy co-evolution and environment training.

Why Use and Key Benefits

This repository offers immense value for anyone interested in the latest advancements in agentic AI. Key benefits include:

Comprehensive Resource: It provides a single, organized location for a vast amount of research, making it easier to navigate the complex landscape of self-evolving agents.
Structured Taxonomy: The content is categorized into clear sections, following a comprehensive taxonomy that helps users understand the different facets and approaches within self-evolution.
Up-to-Date Information: The repository is continuously updated, ensuring access to the newest papers, benchmarks, and projects as the field evolves.
Foundation for Research: Researchers can leverage this collection to identify key trends, understand existing methodologies, and pinpoint areas for future exploration.
Practical Applications: Beyond theoretical papers, it links to open-source projects and real-world applications, bridging the gap between academic research and practical implementation in areas like automated scientific discovery and autonomous software engineering.

Installation

Awesome-Self-Evolving-Agents is a curated list of resources, not a software library that requires installation. To "use" this repository, simply clone it or browse its contents directly on GitHub. All resources, including papers, benchmarks, and links to open-source projects, are accessible through the repository's markdown files.

git clone https://github.com/XMUDeepLIT/Awesome-Self-Evolving-Agents.git
cd Awesome-Self-Evolving-Agents

You can then explore the README.md file and linked documents to delve into the various categories of self-evolving agents.

Examples

The repository is structured to provide clear examples and references across its various sections:

Research Papers: Find extensive lists of papers categorized by evolution mechanism, such as "Inference-Based Evolution" (e.g., Self-consistency, Self-refine) and "Training-Based Evolution" (e.g., Self-instruct, Self-play fine-tuning).
Benchmarks: Discover benchmarks like GPQA for scientific reasoning, GSM8K for mathematical reasoning, and SWE-bench for software engineering, each with links to their respective projects and papers.
Open Source Libraries: Explore foundational agent orchestration tools like LangGraph and AutoGen, as well as libraries for distributed training, post-training, and efficient fine-tuning.
Applications: See real-world implementations in "Automated Scientific Discovery" (e.g., Agon, The AI Scientist) and "Autonomous Software Engineering" (e.g., SWE-agent, Devin).

Each entry typically includes a direct link to the paper or project, facilitating easy access to the original work.