FermiLink: An Agent Harness for Multidomain Scientific Simulations

Abstract Artificial intelligence (AI) agent harnesses have been developed for autonomous scientific simulations, but most current agent harnesses are tested on a single or a small set of software packages. Herein, FermiLink, an extensible open-source agent harness, is introduced for multidomain scientific simulations. Its key design principle is the separation of package knowledge bases from simulation workflows, so that simulation workflows in FermiLink, from figure-level simulations to full-paper-level research on high-performance computing clusters, operate uniformly among supported packages via a four-layer progressive disclosure mechanism. This agent harness is specifically optimized for high-performance computing (HPC) clusters, enabling multiday, multistep scientific simulations in the absence of human intervention. Using OpenAI Codex as the agent provider, the capabilities of FermiLink are demonstrated across approximately 50 scientific software packages spanning nine research domains from chemistry to engineering. Systematic benchmarks on 132 real-world figure-level reproduction tasks with 44 packages show that FermiLink reproduces 74 (56.1%) of published figures with simulations, among which 30 (22.7%) achieve high-fidelity agreement and 35 reach qualitative agreement with the target figures. A smaller set of human expert-guided reproduction benchmarks with 10 packages in physics and chemistry further highlights the importance of expert insights for improving the simulation fidelity. Beyond reproduction, a single-blinded study demonstrates that FermiLink can produce research-grade results on unpublished polariton physics problems when provided with sufficiently detailed research objectives. Overall, FermiLink provides a scalable research infrastructure that may accelerate the path from scientific questions to computational results across diverse domains.