[2603.10062] Multi-Agent Memory from a Computer Architecture Perspective: Visions and Challenges Ahead

Computer Science > Hardware Architecture arXiv:2603.10062 (cs) [Submitted on 9 Mar 2026 (v1), last revised 30 Mar 2026 (this version, v2)] Title:Multi-Agent Memory from a Computer Architecture Perspective: Visions and Challenges Ahead Authors:Zhongming Yu, Naicheng Yu, Hejia Zhang, Wentao Ni, Mingrui Yin, Jiaying Yang, Yujie Zhao, Jishen Zhao View a PDF of the paper titled Multi-Agent Memory from a Computer Architecture Perspective: Visions and Challenges Ahead, by Zhongming Yu and 7 other authors View PDF HTML (experimental) Abstract: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. Subjects: Hardware Architecture (cs.AR); Artificial Intelligence (cs.AI); Multiagent Systems (cs.MA) Cite as: arXiv:2603.10062 [cs.AR]   (or arXiv:2603.10062v2 [cs.AR] for this version)   https://doi.org/10.48550/arXiv.2603.10062 Focus to learn more arXiv-issued DOI via DataCite Submission history From: Zhongming Yu [view email] [v1] Mon, 9 Mar 2...