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.
Figures & tables
Figure 1. Two fundamental multi-agent memory architectures for managing growing context complexity: shared memory and distributed memory. Overview graphic motivating multi-agent memory challenges.
Figure 2. Agent memory hierarchy and protocol framing. Diagram of agent I/O, cache, and memory layers with protocol links.
Figure 3. Consistency model comparison from traditional memory architecture to multi-agent memory. Comparison chart of memory consistency models.
Beijing Advanced Innovation Center for Future Blockchain and Privacy Computing, School of Artificial Intelligence, Beihang University · Beijing academy of blockchain and edge computing · National University of Defense Technology +1