# Position: Episodic Memory is the Missing Piece for Long-Term LLM Agents

**Authors:** Mathis Pink, Qinyuan Wu, Vy Ai Vo, Javier Turek, Jianing Mu, Alexander Huth, Mariya Toneva  
**arXiv:** [2502.06975](https://arxiv.org/abs/2502.06975)  
**Date:** 2025-02-10  
**Summary page:** https://www.summarizepaper.com/en/paper/2502.06975v1/

## Key points

- Authors discuss the evolution of Large Language Models (LLMs) into fully-fledged agents operating in dynamic environments
- The challenge faced by LLMs is the lack of episodic memory for adaptive and context-sensitive behavior
- Proposal to integrate episodic memory framework with five essential properties for long-term LLM agents
- Emphasis on concentrated focus on episodic memory to drive development of long-term agents
- Roadmap outlined to support all five properties of episodic memory for enhanced efficiency and effectiveness

## Summary

In their paper titled "Position: Episodic Memory is the Missing Piece for Long-Term LLM Agents," authors Mathis Pink, Qinyuan Wu, Vy Ai Vo, Javier Turek, Jianing Mu, Alexander Huth, and Mariya Toneva discuss the evolution of Large Language Models (LLMs) from text-completion tools to fully-fledged agents operating in dynamic environments. The  these LLMs face is the  Drawing inspiration from biological systems that utilize episodic memory for single-shot learning of instance-specific contexts, the authors propose an  for LLM agents. This framework is centered around five essential properties of episodic memory that enable adaptive and context-sensitive behavior. The authors argue that with existing research efforts already touching on some aspects of these properties, now is the opportune time for a concentrated focus on episodic memory to drive the development of long-term agents. To this end, they outline a roadmap that integrates various research directions with the goal of supporting all five properties of episodic memory. By doing so, they aim to enhance the efficiency and effectiveness of long-term LLM agents in navigating complex and evolving environments. Overall, this position paper emphasizes the significance of incorporating  to facilitate continual learning and adaptation in real-world scenarios. Through a strategic approach outlined in their roadmap, the authors advocate for a more comprehensive understanding and implementation of episodic memory to propel the advancement of long-term LLM agents towards greater cognitive capabilities and performance.