EvolveR

EvolveR is a framework designed to enable LLM agents to self-improve through a complete, closed-loop experience lifecycle. This repository contains the model weights introduced in the paper EvolveR: Self-Evolving LLM Agents through an Experience-Driven Lifecycle.

Resources

• Paper: EvolveR: Self-Evolving LLM Agents through an Experience-Driven Lifecycle
• Code: https://github.com/Edaizi/EvolveR

Description

Current Large Language Model (LLM) agents show strong performance in tool use but often lack the capability to systematically learn from their own experiences. EvolveR addresses this by introducing a lifecycle comprising:

1. Offline Self-Distillation: Synthesizing interaction trajectories into a structured repository of abstract, reusable strategic principles.
2. Online Interaction: Task interaction guided by retrieved distilled principles to guide decision-making and accumulate behavioral trajectories.

This loop employs a policy reinforcement mechanism to iteratively update the agent based on its performance.

Citation

@article{wu2025evolver,
  title={EvolveR: Self-Evolving LLM Agents through an Experience-Driven Lifecycle},
  author={Wu, Rong and Wang, Xiaoman and Mei, Jianbiao and Cai, Pinlong and Fu, Daocheng and Yang, Cheng and Wen, Licheng and Yang, Xuemeng and Shen, Yufan and Wang, Yuxin and Shi, Botian},
  journal={arXiv preprint arXiv:2510.16079},
  year={2025}
}