This repository provides a curated survey and paper list focused on efficiency-guided design for Large Language Model (LLM) agents. It addresses the critical need for efficiency in LLM agents, which has often been overlooked in favor of capability advancements, making it an invaluable resource for researchers, engineers, and power users seeking to understand and develop practical, deployable LLM agent systems. The primary benefit is a structured overview of research in memory, tool learning, and planning, enabling quick access to representative work and insights into efficiency improvements.

How It Works

The project categorizes and surveys research papers based on three core components of efficiency-guided LLM agents: memory, tool learning, and planning. Memory-related work is organized by construction, management, and access processes. Tool learning papers are grouped into selection, calling, and integrated reasoning. Planning research focuses on methods that enhance overall agent efficiency by minimizing unnecessary actions and shortening execution trajectories, offering a systematic approach to understanding efficiency drivers in agent design.

Quick Start & Requirements

This repository serves as a curated survey and paper list, not a runnable software project. It does not provide installation instructions or direct execution requirements.

Highlighted Details

• Offers a comprehensive, categorized paper list covering memory (Working, Latent, External, Graph-based, Hierarchical, Multi-Agent), tool learning (Selection, Calling, Integrated Reasoning), and planning (Single-Agent, Multi-Agent Collaborative Efficiency).
• Each paper entry is linked to its research context, with notes on how it connects to agent efficiency.
• Includes a section for related survey papers, providing complementary perspectives on LLM agents, memory, tool learning, and planning.
• The project actively welcomes contributions for adding papers, fixing links, or improving categorization.

Maintenance & Community

Contributions are welcomed via GitHub issues or pull requests to enhance the paper list and categorization. Specific community channels (e.g., Discord, Slack) or details on core maintainers are not provided in the README.

Licensing & Compatibility

No specific software license is mentioned for the repository's content or curation. The primary output is a list of research papers, each governed by its own license.

Limitations & Caveats

As a curated list of research papers, this repository does not offer executable code or direct deployment capabilities. The information reflects the state of research as compiled by the authors, and specific paper implementations may vary in maturity and availability. The timeline for a "revised version" (April/May) may have passed.