E-mem: Multi-agent based Episodic Context Reconstruction for LLM Agent Memory
Kaixiang Wang, Yidan Lin, Jiong Lou, Zhaojiacheng Zhou, Bunyod Suvonov, Jie Li
Why It Matters
What makes this one worth your time
Improving memory handling in LLMs can significantly enhance their reasoning capabilities, making them more effective for complex problem-solving tasks.
E-mem enhances LLM agent memory by reconstructing episodic contexts using a multi-agent system.
Summary
The paper introduces E-mem, a framework for episodic context reconstruction in large language model agents, aiming to enhance System 2 reasoning by maintaining uncompressed memory contexts through a multi-agent architecture. The approach is inspired by biological engrams and involves assistant agents handling local reasoning and a master agent for global planning. The framework reportedly surpasses the state-of-the-art in the LoCoMo benchmark in terms of F1 score and token cost efficiency.
Key contributions
- Proposal of E-mem framework for episodic context reconstruction.
- Introduction of a multi-agent system to maintain uncompressed memory contexts.
- Demonstration of improved performance on the LoCoMo benchmark.
Notable insights
- The use of a heterogeneous hierarchical architecture with assistant and master agents for memory management.
- Inspiration from biological engrams to maintain contextual integrity in memory processing.
Possible limitations
- Not stated in the abstract
Abstract
arXiv:2601.21714v4 Announce Type: replace Abstract: The evolution of Large Language Model (LLM) agents towards System~2 reasoning, characterized by deliberative, high-precision problem-solving, requires maintaining rigorous logical integrity over extended horizons. However, prevalent memory preprocessing paradigms suffer from destructive de-contextualization. By compressing complex sequential dependencies into pre-defined structures (e.g., embeddings or graphs), these methods sever the contextual integrity essential for deep reasoning. To address this, we propose E-mem, a framework shifting from Memory Preprocessing to Episodic Context Reconstruction. Inspired by biological engrams, E-mem employs a heterogeneous hierarchical architecture where multiple assistant agents maintain uncompressed memory contexts, while a central master agent orchestrates global planning. Unlike passive retrieval, our mechanism empowers assistants to locally reason within activated segments, extracting context-aware evidence before aggregation. Evaluations on the LoCoMo benchmark demonstrate that E-mem achieves over 54\% F1, surpassing the state-of-the-art GAM by 7.75\%, while reducing token cost by over 70\%.