Introducing PegasusAI Studio: A new way of Artificial Intelligence Assisted Workflow Management on National Cyberinfrastructure

Abstract

Workflows are a key technology for enabling complex scientific computations. They capture the interdependencies between processing steps in data analysis and simulation pipelines as well as the mechanisms to execute those steps reliably and efficiently. Workflows can capture complex processes, promote sharing and reuse, and also provide provenance information necessary for the verification of scientific results and scientific reproducibility. PegasusAI (https://pegasus-ai.org/) is an open, AI-infused workflow management framework that extends the Pegasus Workflow Management System to enable intelligent orchestration of scientific workflows across the computing continuum, spanning edge resources, cloud platforms, and high-performance computing (HPC) systems. The project integrates machine learning into workflow planning, resource provisioning, execution monitoring, anomaly detection, performance prediction, and adaptive workflow management, while providing role-based, human-in-the-loop interfaces for scientists, cyberinfrastructure facilitators, and CI operators. PegasusAI also produces curated, AI-ready workflow execution datasets and reusable models to accelerate research in AI for scientific workflows and cyberinfrastructure operations. By combining trusted workflow technologies with AI-driven decision support, PegasusAI aims to improve the resilience, efficiency, and accessibility of distributed scientific computing. As part of PegasusAI we have been working on a hosted deployment called PegasusAI Studio (studio.pegasus-ai.org), that allows users to build, run, monitor, and debug Pegasus workflows with AI assistance — in one persistent browser workspace, from scientific intent to a running, inspectable run. PegasusAI Studio also comes with an in-built marketplace that features a variety of representative workflows from different science domains that users can use as a starting reference point for their workflows. Users can also generate workflows from the chat interface by combining their own prompts with workflow-specific skills, sent to whichever LLM backend they have connected. PegasusAI studio leverages core pegasus monitoring and debugging tools augmented with AI capabilities, including new web-based monitoring tools that give real time feedback to the users about their running workflows. For workflow execution, the studio seamlessly also connects with a variety of national Cyber Infrastructure resources such as ACCESS, PATh Facility, the OSPool, Fabric and Chameleon. PegasusAI studio is under active development with its public release scheduled for November 2026. In this poster, we present the vision and architecture of PegasusAI Studio and give a preview of its capabilities. Pegasus AI is a joint collaboration between Information Sciences Institute (ISI) at University of Southern California, Renaissance Computing Institute (RENCI) at University of North Carolina-Chapel Hill, University of Tennessee at Knoxville and University of Massachusetts, Amherst.