This repository provides a comprehensive survey of LLM-based autonomous agents, targeting researchers and practitioners in AI. It systematically analyzes the construction, applications, and evaluation of these agents, offering a valuable resource for understanding this rapidly evolving field.

How It Works

The survey categorizes LLM-based autonomous agents by their core components: profile, memory, planning, and action modules. It then explores their applications across natural sciences, social sciences, and engineering, and discusses various evaluation strategies, including subjective and objective methods. The work highlights the evolution of AI capabilities from machine learning to LLMs and now to agents, introducing "mechanism engineering" alongside parameter and prompt engineering.

Quick Start & Requirements

This repository is a survey paper and does not contain executable code for building or running agents. The primary content is the survey paper itself, available via the provided arXiv link.

Highlighted Details

• First published survey paper specifically on LLM-based autonomous agents.
• Includes detailed tables categorizing agents by construction strategy, application domain, and evaluation methods.
• Features new figures illustrating planning approaches and the evolution of AI capabilities.
• Maintains an interactive table with additional papers, allowing for sorting and filtering.

Maintenance & Community

The survey is maintained by Lei Wang, Chen Ma, and Xueyang Feng. Contributions are welcomed via pull requests, issues, or email. Contact information for Lei Wang and Xu Chen is provided.

Licensing & Compatibility

The repository itself does not specify a license. The survey paper is available on arXiv, and its publication in Frontiers of Computer Science suggests adherence to academic publishing standards.

Limitations & Caveats

As a survey paper, this repository does not provide functional code or tools. Its primary value is in its comprehensive overview and categorization of existing research in LLM-based autonomous agents.