Epeditor: A scalable platform for large-ensemble building energy simulation enabling agent-assisted automation

The data scale of energy simulation keeps increasing, particularly in urban building energy modeling (UBEM) and uncertainty analysis, where thousands of EnergyPlus runs are often required to explore complex parameter spaces and building variability. To simplify these data-intensive workflows, recent studies have introduced AI agents for end-to-end simulation tasks. However, in practice, both AI-assisted and conventional batch workflows remain constrained by unstable field inference and search bias, inefficient large-ensemble simulation execution, and cumbersome data processing. To address these problems, this study introduces Epeditor, a cloud-deployable simulation infrastructure for large-ensemble building energy analysis for both AI and human users. It combines three improvements on current workflows: a knowledge-assisted parameterization with more accurate and stable field search in IDF files; a simulator that improves parallel execution efficiency through Operating-System-Level deployment; and a data management system that enables memory-safe aggregation of large simulation inputs and outputs. The packaged Agent Skill allows AI agents to execute fully automated dataset generation, batch simulation, surrogate modeling, and interpretation through an auditable workflow. Epeditor is evaluated through a agent benchmarks with metrics for Artificial General Intelligence, and two case studies are introduced to show its ability in research and design practices including an AI-assisted surrogate-modeling case, and design-studio practices. The pipeline exhibits 99.98% robustness in IDF generation, up to four-fold acceleration in high-parallel simulation, and 76.28 ( F 1 ) in IDF field inference. These findings indicate that Epeditor can lower the technical and computational barriers to large-ensemble EnergyPlus simulation, supporting more accessible, reproducible, and AI-ready building energy analysis.