Artificial intelligence news.
Straight from the original announcements.

PegasusAI Studio Connects AI Agents to Real Scientific Computing

November 12, 2026, 21:39By the AI News Bites Editorial TeamSource: USC Viterbi (AI)
PegasusAI Studio Connects AI Agents to Real Scientific Computing
Image: USC Viterbi (AI)

PegasusAI Studio, an open-source platform from USC and research partners, is designed to help AI agents build, run, monitor and debug workflows on real scientific computing infrastructure. The platform became publicly available through a science gateway on November 12, 2026.

Key points
  • PegasusAI Studio is an open-source platform for AI-assisted scientific workflows.
  • It connects agents to workflow systems, schedulers, data repositories, clouds and HPC systems.
  • Tests included 350 gravitational-wave jobs, 127 EEG jobs and 55 Event Horizon Telescope jobs.
  • Agents operate under the scientist's identity and existing approval controls.
  • The platform is publicly available through studio.pegasus-ai.org.

From code generation to scientific execution

Artificial intelligence can write code, but scientific computing also requires data movement, specialized software environments, job coordination and troubleshooting across clusters, research clouds, supercomputers and distributed systems. PegasusAI Studio is designed to connect AI agents to those existing resources rather than recreate the underlying infrastructure inside the AI layer.

The platform builds on the Pegasus Workflow Management System and the HTCondor Software Suite. It gives agents current information about workflow systems and computing environments, tools for managing and running workflows, and structured details about active or failed jobs.

That information can include job states, scheduler holds, logs, data-staging details and diagnostic reports. According to USC Viterbi, agents can use this context to help diagnose failures, repair workflows or software environments, and resubmit jobs.

Tests across scientific applications

The researchers evaluated PegasusAI Studio with applications from several areas of NSF-supported computational science. In weather-radar workflows, it helped identify and repair infrastructure and configuration problems while modernizing legacy scientific software.

In a gravitational-wave analysis, an AI-written workflow was converted into an interactive Studio application and successfully executed 350 jobs. An EEG analysis, using a different AI agent and an open-weight model, completed 127 jobs across data from 24 subjects after recovering from an execution-environment problem.

The team also used Event Horizon Telescope data to reconstruct images of the black hole in galaxy M87. Two initial attempts failed because the required software containers had been built incorrectly. After Studio passed failure details to an AI debugging agent, the next attempt completed all 55 jobs without a failure and produced the reconstructed images.

Keeping AI actions accountable

PegasusAI Studio is designed to keep AI agents within the identity, security and resource controls already used by scientific facilities. Agents operate under the identity and permissions of the scientist using them, while submitted jobs remain attributable to that researcher. Higher-risk actions and workflow submissions remain subject to user approval controls.

“The agent should work through the same trusted systems as the scientist,” said Mats Rynge, one of the platform's main architects, according to USC Viterbi.

Studio can work with different AI models and providers, including commercial models, research-hosted services and open models. This lets institutions choose according to their cost, privacy and data-governance requirements.

Public access and education

Researchers can use PegasusAI Studio through the public science gateway at studio.pegasus-ai.org. Sign-in uses CILogon with institutional credentials or another supported provider. Accounts are usually approved within one business day, after which a workspace is set up automatically and a guided tour appears at first login. Institutions can also deploy Studio in their own computing environments.

The project is an NSF-funded collaboration among the University of Southern California, the Renaissance Computing Institute, the University of Massachusetts Amherst and the University of Tennessee, Knoxville. It is supported under NSF Award #2513101.

Studio is also being tested in education. Michela Taufer and Sai Swaminathan piloted it in an EECS data engineering course at UTK in spring 2026, with full deployment planned there in spring 2027. The team is also using it in hackathons where students build soil-moisture workflows and examine how AI agents share control with human users.

Related news

AWS shows how to route coding agent requests with NeMo Switchyard
For developersNovember 13, 2026, 23:30

AWS shows how to route coding agent requests with NeMo Switchyard

AWS has published a sample deployment that routes coding agent requests between models on Amazon Bedrock using NVIDIA NeMo Switchyard and LiteLLM. The approach measures…

USC ISI to Present AI, Workflow and Cybersecurity Research at SC26
ResearchNovember 13, 2026, 22:43

USC ISI to Present AI, Workflow and Cybersecurity Research at SC26

Researchers from the University of Southern California’s Information Sciences Institute will present work on scientific workflow automation, artificial intelligence for…

Cloudflare open-sources Kitesurf browser engine for AI agents
For developersNovember 13, 2026, 21:20

Cloudflare open-sources Kitesurf browser engine for AI agents

Cloudflare has open-sourced Kitesurf, a stateless browser for AI agents that runs entirely on Cloudflare Workers. Developers can now inspect the code, run it locally…