AutoMOOSE: Multi-Agent AI Framework Automates Phase-Field Simulations

The newly created open-source multi-agent AI framework, AutoMOOSE, aims to streamline phase-field simulations essential for connecting thermodynamics and kinetics to microstructural changes. This framework, outlined in an arXiv paper (2603.20986), manages the entire simulation process from a single natural-language command, simplifying the complexities often associated with multiphysics frameworks like MOOSE that usually necessitate expert knowledge. AutoMOOSE features six distinct agents: Architect, Input Writer, Runner, Reviewer, Visualization, and a physics-based Skeptic. These agents are responsible for generating, executing, analyzing, and rigorously testing simulations against conservation laws and scaling relations. Validation occurred in two areas: non-conserved copper grain growth via Allen-Cahn dynamics and conserved Fe-Cr spinodal decomposition through Cahn-Hilliard dynamics. AutoMOOSE effectively completed a 25-task grain-growth benchmark, showcasing its potential to facilitate complex materials science modeling and enhance research in materials design.