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

TL;DR

This paper frames multi-agent memory as a computer architecture challenge, proposing a layered hierarchy and identifying key gaps like cache sharing and access control for scalable, reliable systems.

Contribution

It introduces a novel architectural perspective on multi-agent memory, emphasizing hierarchy design and protocol gaps, especially memory consistency challenges.

Findings

Proposes a three-layer memory hierarchy for multi-agent systems

Identifies critical protocol gaps in cache sharing and access control

Highlights memory consistency as the most pressing open challenge

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.