XMUDeepLIT/Awesome-Self-Evolving-Agents

What it solves

This repository compiles research on AI agents that can improve themselves (self-evolution). It organizes hundreds of papers into a taxonomy (model-centric, environment-centric, co-evolution) to help researchers find relevant work, understand research trends, and discover benchmarks and open-source libraries.

How it works

The README works as an index. It groups papers into three main branches: Model-Centric Self-Evolution (improving the model), Environment-Centric (improving the agent's interaction with external knowledge) and Model-Environment Co-Evolution. Each branch has subcategories like inference-based evolution, training-based, RAG, experience evolution, etc. It also lists related surveys, benchmarks, open-source libraries, and applications. Each entry links to the original paper or resource.

Who it's for

Researchers, students, and developers working on AI agents, LLMs, self-improving systems, or lifelong learning. It's especially useful for anyone looking for a structured starting point or literature review on self-evolving agents.

Highlights

• Comprehensive taxonomy covering three major evolution paradigms.
• Continuously updated with new papers and news.
• Includes links to benchmarks, libraries, and applications.
• Welcomes contributions via PRs and issues.