HPC workflow automation
An agent connects Slurm, ADIOS2, Jarvis, and a Gray–Scott simulation into one automated HPC workflow.
IOWARP / DEMONSTRATIONS
For anyone who wants to see the software run: recordings of CLIO software on real scientific data and systems, made by the IOWarp team. The 2026 series follows the full stack, from CUDA kernels to DOE laboratory workflows; each entry states what its recording will show. The 2025 archive holds the first agent workflows on EarthScope, LAMMPS, ParaView, ADIOS2, and Slurm.
These entries are placeholders, not previews. Each video appears here when it is published.
A clean machine to a working scientific agent: install CLIO Kit, connect an MCP client, and open a real HDF5 dataset.
The NDP workflow rebuilt on the current releases: catalog search, staging, profiling, and plotting of GNSS data.
An agent submits and watches Slurm jobs, then reads the Darshan I/O profile to explain what the job did to the file system.
CUDA kernels in CLIO Core write GPU-resident data into IOWarp and stage it through host memory and NVMe.
One Jarvis pipeline description deploys the same CLIO Core configuration on DeltaAI at NCSA and Aurora at Argonne.
A materials question becomes a synthesis of the supplied literature, a map of the available lab, and a reviewed research plan.
CLIO Agent resumes a data investigation days later and recalls what it learned and where the data lives.
CLIO Agent investigating provenance with the SPOTTER-AI team at Argonne, and CLIO Coder fleets working with the PORTUS team at PNNL.
CLIO Coder spreads bounded engineering tasks across workers on cluster nodes and returns reviewable run receipts.
Recorded between September and November 2025. Select a preview to load the video player.
An agent connects Slurm, ADIOS2, Jarvis, and a Gray–Scott simulation into one automated HPC workflow.
An agent examines EarthScope seismograph data found through the National Data Platform.
An agent answers questions about data produced by a LAMMPS molecular dynamics simulation.
ParaView exposed through an MCP server so an agent can build and adjust a scientific visualization.
An agent works through a scientific analysis with the IOWarp MCP servers, from data access to a finished plot.
An agent reads and inspects ADIOS2 simulation output through the IOWarp MCP servers.
A workflow deployed reproducibly, with the agent’s activity streamed to a dashboard for inspection.

The six-stage materials workflow that the demonstrations aim to cover: from a tomography dataset to reconstruction, segmentation, storage, and visualization. Every handoff between tools is part of the scientific work.
The materials science storyWhat changed since
The 2025 recordings show agents calling individual MCP tools. The 2026 series shows the layers underneath and around them: GPU offload in CLIO Core, deployment on national systems, memory that persists across sessions, and agents working inside DOE laboratory projects.
See the platform