Agent-Kernel
A multi-agent system framework for building large-scale, LLM-driven social simulations, supporting agents added or removed at runtime, live intervention during a run, and verification of agent behavior against model outputs.
Agent-Kernel is designed to let researchers add and remove LLM-based agents while a simulation runs, scale to large numbers of agents, intervene in a simulation in real time, and verify agent behavior and model outputs, while reusing code across scenarios through its Controller module.
It has been applied to simulations such as a recreation of the "Universe 25" population-density experiment, a high-fidelity model of a university campus for studying pedestrian flow and resource allocation, and an interactive hospital arena for benchmarking collective intelligence in clinical workflows.
It is written in Python and includes an optional Society-Panel component that can be started alongside a simulation.