As LLM agents evolve into collaborative multi-agent systems, their memory requirements grow rapidly in complexity. This position paper frames multi-agent memory as a computer architecture problem. We distinguish shared and distributed memory paradigms, propose a three-layer memory hierarchy (I/O, cache, and memory), and identify two critical protocol gaps: cache sharing across agents and structured memory access control. We argue that the most pressing open challenge is multi-agent memory consistency. Our architectural framing provides a foundation for building reliable, scalable multi-agent systems.
@article{arxiv.2603.10062,
title = {Multi-Agent Memory from a Computer Architecture Perspective: Visions and Challenges Ahead},
author = {Zhongming Yu and Naicheng Yu and Hejia Zhang and Wentao Ni and Mingrui Yin and Jiaying Yang and Yujie Zhao and Jishen Zhao},
journal= {arXiv preprint arXiv:2603.10062},
year = {2026}
}