Agent-Kernel: A MicroKernel Multi-Agent System Framework for Adaptive Social Simulation Powered by LLMs

TL;DR

Agent-Kernel tackles the rigidity of existing MAS frameworks by introducing a society-centric modular microkernel that decouples cognitive agents from external environments and actions. The framework provides a five-module core (Agent, Environment, Action, Controller, System) and a plugin-based ecosystem, enabling runtime adaptability, global configurability, deterministic execution, and strong reusability via a Database-per-Plugin model. Demonstrations on Universe 25 and the Zhejiang University Campus Life showcase the system's ability to handle evolving populations and large-scale, heterogeneous agent populations with detailed performance and behavioral metrics. The work highlights practical implications for scalable, reusable, and reliable social simulations powered by LLM-driven agents and offers avenues for community sharing through SocietyHub.

Abstract

Multi-Agent System (MAS) developing frameworks serve as the foundational infrastructure for social simulations powered by Large Language Models (LLMs). However, existing frameworks fail to adequately support large-scale simulation development due to inherent limitations in adaptability, configurability, reliability, and code reusability. For example, they cannot simulate a society where the agent population and profiles change over time. To fill this gap, we propose Agent-Kernel, a framework built upon a novel society-centric modular microkernel architecture. It decouples core system functions from simulation logic and separates cognitive processes from physical environments and action execution. Consequently, Agent-Kernel achieves superior adaptability, configurability, reliability, and reusability. We validate the framework's superiority through two distinct applications: a simulation of the Universe 25 (Mouse Utopia) experiment, which demonstrates the handling of rapid population dynamics from birth to death; and a large-scale simulation of the Zhejiang University Campus Life, successfully coordinating 10,000 heterogeneous agents, including students and faculty.

Figures (11)

• Figure 1: Comparison between pipeline architecture, layered architecture and modular microkernel architecture. The , , and symbols represent the agent, environment and action module, respectively; and stands for the coupling of the environment and action modules under the agent.
• Figure 2: Simulation of the Universe 25 experiment. (a) The initialization stage, featuring an equal sex ratio of four pairs (4 males, 4 females). (b) Mating behavior, showing the male mouse pursuing the female. (c) A representative scenario of collective behaviors. (d) The population growing to 111 individuals.
• Figure 3: Simulation of the Zijingang campus, Zhejiang University. (a) Panoramic view of the campus. (b) Several agents discussing in a laboratory. (c) The dining scenario in a canteen. (d) Conversation between two agents.
• Figure 4: Illustration of the modular microkernel architecture of Agent-Kernel. represents a module component, and represents a plugin.
• Figure 5: Software Design of the Agent-Kernel framework.