Summary

This repository, TsinghuaC3I/Awesome-Memory-for-Agents, functions as a meticulously curated bibliography for researchers and practitioners in the field of AI agents. It addresses the critical challenge of memory in language agents by systematically cataloging academic papers, surveys, benchmarks, and related projects. The primary benefit is providing a structured, navigable resource that accelerates understanding and innovation in developing agents capable of retaining and utilizing information across tasks and interactions, targeting anyone from academic researchers to engineers building sophisticated AI systems.

How It Works

The core of this resource is a taxonomy that classifies agent memory based on its persistence and curation mechanism. Memory is broadly divided into Short-Term Memory, which is transient and confined to a single task's context window, and Long-Term Memory, designed for persistence across multiple tasks. Long-Term Memory is further differentiated into "Experience," which involves knowledge explicitly validated or learned from task outcomes (successes or failures), and "Memory," which encompasses information stored without direct reference to task outcomes. This structured approach directly maps to three primary application scenarios that organize the vast collection of papers: Personalization (e.g., user profiles, interaction history), Learning from Experience (e.g., skill acquisition, trajectory optimization), and Long-horizon Agentic Tasks (e.g., managing intermediate results, reasoning traces).

Quick Start & Requirements

This repository is a curated list of research papers and does not offer software for installation or execution. It serves as a comprehensive reference guide for the academic and product landscape of agent memory.

Highlighted Details

• Features a robust taxonomy distinguishing Short-Term Memory from Long-Term Memory, further subdividing the latter into "Experience" and general "Memory."
• Organizes research papers and projects across three key application domains: Personalization, Learning from Experience, and Long-horizon Agentic Tasks, providing clear use-case mappings.
• Includes an extensive and up-to-date collection of academic papers (with dates ranging from 2020 to late 2025), relevant surveys, benchmarks for evaluating agent memory, and a list of related products and projects in the domain.
• Highlights recent advancements and emerging trends, with a significant number of entries from 2025, indicating a focus on cutting-edge research.

Maintenance & Community

The paper list is actively maintained by Hongyi Liu, Yu Fu, and Kaiyan Zhang, with contributions acknowledged from Yuxin Zuo, Che Jiang, Guoli Jia, Yuru Wang, Kaikai Zhao, Yuchen Fan, Zhenzhao Yuan, Kai Tian, and Weizhi Wang, suggesting a collaborative effort within the research community.

Licensing & Compatibility

The README does not specify a license for this curated list of papers. Users should consult the individual licenses of the papers and projects referenced for any compatibility concerns regarding commercial use or closed-source integration.

Limitations & Caveats

As a bibliography, this repository does not provide executable code, tools, or direct access to the full text of the papers. Its scope is strictly limited to memory mechanisms within language agents, excluding broader agent development topics. The dynamic nature of AI research means the list requires continuous updates to maintain its comprehensiveness and relevance.