CLIO · CONTEXT LAYER FOR I/O
IOWarp builds CLIO, open-source software that puts AI agents to work on your scientific data and leaves a record your people can check. It runs on the systems you already operate, from a workstation to a national facility. In AI-driven science, context is part of the system.
THE CLIO SUITE
Choose the product closest to your work. CLIO Agent and CLIO Coder serve people working on scientific data and scientific software; CLIO Kit supplies the scientific tools both share, and CLIO Core runs the context hierarchy underneath.
Flagship Scientific data workspace
For scientists, lab teams, and facility data managers.
Work with files, live streams, visualizations, shared web pages, and LaTeX papers in the desktop app or a browser, on local, cloud, or lab models.
curl -fsSL https://clio.iowarp.ai/install.sh | bash Flagship Meta-harness for scientific software
For developers, computational scientists, and research software engineers.
Bring every model you have behind shared project guidance, inspectable checks, measured research campaigns, and a blind model arena.
curl -fsSL https://coder.iowarp.ai/install.sh | sh Flagship Scientific tools and workflows
For any MCP client: CLIO Agent, CLIO Coder, Claude Code, Codex, Cursor, and VS Code.
Select scientific MCP servers, skills, and workflow plugins for HDF5, Parquet, ADIOS2, Slurm, ParaView, the National Data Platform, and more.
uv tool install clio-kitInfrastructure Tiered context runtime
For HPC developers, facilities, and pipelines spanning memory and storage tiers.
Assimilate scientific data with its metadata, place it across GPU memory, DRAM, NVMe, and parallel storage, and query it from Python or C++.
pip install iowarp-coreIOWarp will launch the full CLIO stack at SC26 in Chicago, November 15–20, 2026. IOWarp at SC26 →
WHAT IS THIS?
IOWarp is the research and engineering team at the Gnosis Research Center, Illinois Tech, that builds CLIO. CLIO, the Context Layer for I/O, is an open-source software stack that connects AI agents to scientific data and the systems that hold it.
It matters because scientific data is only useful with its meaning attached. Point an agent at an HDF5 file or a Slurm queue and it needs the right data, memory that outlasts the session, and a path from a laptop to a national system. CLIO supplies each of those, so an organization can bring AI into its science and hold every change to its own checks.
Adoption is incremental. CLIO Agent, CLIO Coder, CLIO Kit, and CLIO Core each install on their own.
From sources and results to a paper someone can check.
Science workflows and domain packages built on the stack.
Tiered devices, I/O adapters, compression, and data connectors.
The Linux I/O path, from asynchronous I/O to kernel modules.
GPU code that reaches the storage hierarchy directly.
ACROSS DISCIPLINES
CLIO began in materials science, expanded to mechanical engineering simulations when neighboring groups saw it run, and now reaches analytical chemistry, federated earth-science catalogs, and DOE Genesis Mission workflows at Argonne and PNNL. Each discipline keeps the same CLIO backbone while swapping in the tools, domain skills, and checks its workflow demands.
01 · Where CLIO started
Additive manufacturing, X-ray computed tomography, segmentation, and next-experiment planning across instruments and clusters.
Printed specimens move from design to tomography scans and fatigue testing, where process history and volumes have to stay linked.
wtf-MS and Materio study-planning plugins, HDF5 and Parquet tools, ParaView rendering, and Globus data assimilation.
CLIO Agent keeps study questions and observations tied to tool calls; CLIO Core holds the shared part record across stages.
02 · Adopted for simulation
Structural and topology-optimization campaigns where engineers saw the materials agents and adapted them to finite-element models.
Engineers configure Abaqus and Tosca input decks, sweep parameters on cluster queues, inspect stress fields, and compare simulation runs.
Abaqus model-interpretation skills, Slurm and Lmod cluster servers, ParaView field inspection, and numerical check gates.
CLIO Coder verifies code and configuration edits against declared checks; CLIO Agent records parameter choices and figures.
03 · Adopted at the bench
Experimental labs working from bench instruments, assay tables, and small-data machine learning without needing a supercomputer.
Measurements stay at the bench while students and researchers organize instrument exports, train and validate predictive models, and preserve continuity when cohorts rotate.
Pandas and Parquet profiling, scripted matplotlib figures, marimo notebooks, and method-validation checks on held-out samples.
CLIO Agent keeps computational steps as auditable as a validated lab method, and CLIO Coder reruns model pipelines from pinned inputs.
04 · Scaled to facilities
Federated geoscience catalogs, simulation monitoring, workflow provenance security, and cross-facility HPC and cloud orchestration.
Agents discover EarthScope stations through the National Data Platform, trace workflow integrity in SPOTTER-AI, and adapt multi-site runs in PORTUS.
NDP, Geo, and Seismology tools, the SPOTTER provenance investigator and watcher, and PORTUS surrogate performance models.
The same MCP tool contracts, bounded agent permissions, sealed run receipts, and tiered CLIO Core placement underneath.
05 · Bring your domain
Any scientific group whose day crosses literature, messy data files, analysis scripts, predictive models, and manuscripts.
Keep physical instruments and scientific judgment with your people, and hand the repetitive reading, data plumbing, code verification, and record-keeping to CLIO.
Pick existing servers from CLIO Kit or add a focused MCP server, a plain-Markdown domain skill, and your project’s own validation checks.
CLIO Agent, CLIO Coder, CLIO Kit, and CLIO Core run on the models and machines your organization already approves.
IN THE FIELD
Select a collaboration to preview its workflow and inspect how CLIO operates alongside working scientists and facilities.
THE LAUNCH
IOWarp at SC26November 15–20, 2026 · McCormick Place, Chicago
IOWarp will launch the full CLIO stack at SC26 in Chicago. The team will present work on simulation monitoring, GPU I/O, storage observability, compressed training data, and agent fleets through workshop presentations, research posters, a conference paper, and an HDF5 birds-of-a-feather session with The HDF Group. The user group will gather around these events, and the Technical Advisory Board will form at the end of November.
Open source
The IOWarp team at the Gnosis Research Center builds CLIO in the open, alongside the scientific teams and facilities that put it to work.
