Papers
Topics
Authors
Recent
Gemini 2.5 Flash
Gemini 2.5 Flash
72 tokens/sec
GPT-4o
61 tokens/sec
Gemini 2.5 Pro Pro
44 tokens/sec
o3 Pro
8 tokens/sec
GPT-4.1 Pro
50 tokens/sec
DeepSeek R1 via Azure Pro
28 tokens/sec
2000 character limit reached

A Survey on the Memory Mechanism of Large Language Model based Agents (2404.13501v1)

Published 21 Apr 2024 in cs.AI
A Survey on the Memory Mechanism of Large Language Model based Agents

Abstract: LLM based agents have recently attracted much attention from the research and industry communities. Compared with original LLMs, LLM-based agents are featured in their self-evolving capability, which is the basis for solving real-world problems that need long-term and complex agent-environment interactions. The key component to support agent-environment interactions is the memory of the agents. While previous studies have proposed many promising memory mechanisms, they are scattered in different papers, and there lacks a systematical review to summarize and compare these works from a holistic perspective, failing to abstract common and effective designing patterns for inspiring future studies. To bridge this gap, in this paper, we propose a comprehensive survey on the memory mechanism of LLM-based agents. In specific, we first discuss ''what is'' and ''why do we need'' the memory in LLM-based agents. Then, we systematically review previous studies on how to design and evaluate the memory module. In addition, we also present many agent applications, where the memory module plays an important role. At last, we analyze the limitations of existing work and show important future directions. To keep up with the latest advances in this field, we create a repository at \url{https://github.com/nuster1128/LLM_Agent_Memory_Survey}.

Overview of Memory Mechanisms in LLM-based Agents

Introduction to Memory in Agents

LLMs have been integrated into a variety of agent-based applications, enhancing the agents' capability to interact, learn, and evolve within their environments. A critical component that differentiates these sophisticated agents from their earlier counterparts is their memory module. The memory empowers agents to recall past interactions, utilize learned experiences, incorporate external knowledge, and consequently, make informed decisions in real-time scenarios.

Memory in agents spans storing interaction history within a session (trial), recalling information from past sessions (cross-trial), and tapping into external databases or using APIs to access contemporary data. These memory sources are utilized in various operational forms, chiefly as textual and parametric memories. Textual memories are directly interpretable and involve sequences or structures of text, while parametric memories integrate information into the model's parameters, offering a more compact and efficient storage method but at the cost of direct interpretability.

The operationalization of memory involves three phases: writing, managing, and reading. These correspond to storing new information, processing and organizing this stored data, and retrieving appropriate memories to inform current interactions, respectively.

Theoretical and Practical Implications of Agent Memory

Theoretically, agent memory aligns with principles from cognitive psychology, suggesting that replicating human-like memory processes in agents could facilitate more natural and effective interactions. Practically, memory supports self-evolution by enabling learning from past mistakes, optimizing exploratory actions, and abstracting knowledge to generalize across different environments.

Agent memory finds applications in multifarious domains such as personal assistants, role-playing, social simulations, and more. These applications benefit distinctly from memory; for instance, personalized interactions in assistant applications rely on recalling user preferences and past interactions, while role-playing might utilize memory to maintain character consistency and backstory.

Challenges and Future Directions

Despite the advances, the implementation and management of memory in LLM-based agents confront several challenges:

  • Scalability: Managing an ever-growing memory, particularly in long-lived agents or complex environments, can be problematic.
  • Efficiency: Textual memory, while interpretable, may not scale efficiently, making parametric memory forms a potentially more viable option in resource-constrained scenarios.
  • Complexity in Management: Ensuring that the memory contains relevant, non-redundant, and up-to-date information requires sophisticated management strategies.

Future research directions may involve developing more advanced parametric memory solutions that balance efficiency with interpretability, exploring multi-agent memory systems where agents share and synchronize memories to enhance collective decision-making, and enhancing the robustness of memory mechanisms to manage the complexity and dynamics of real-world environments effectively.

Conclusion

Memory in LLM-based agents represents a dynamic and crucial area of research that underpins the effectiveness of agent applications in diverse real-world scenarios. Advancements in this area hold the promise of creating more autonomous, intelligent, and versatile agent systems, driving the next generation of artificial intelligence applications.

User Edit Pencil Streamline Icon: https://streamlinehq.com
Authors (9)
  1. Zeyu Zhang (143 papers)
  2. Xiaohe Bo (4 papers)
  3. Chen Ma (90 papers)
  4. Rui Li (384 papers)
  5. Xu Chen (413 papers)
  6. Quanyu Dai (39 papers)
  7. Jieming Zhu (68 papers)
  8. Zhenhua Dong (76 papers)
  9. Ji-Rong Wen (299 papers)
Citations (53)
Youtube Logo Streamline Icon: https://streamlinehq.com