Fuzzball 4.2 gives enterprises one platform to build, train, and serve their own AI 

A new MCP server lets AI agents operate Fuzzball under explicit permission, and the workflows running inside the cluster can now direct additional work themselves.

RENO, Nev., Sept. 03, 2026 (GLOBE NEWSWIRE) -- CIQ, the enterprise software company behind Rocky Linux, today announced Fuzzball 4.2, a release of its turnkey sovereign AI and HPC orchestration platform that helps organizations turn training and inference jobs from fragile infrastructure projects into reliable production workloads. The release also connects the platform to the agentic AI ecosystem, so an AI agent can now operate Fuzzball directly, and the workloads running inside it can orchestrate additional work themselves. Long-running AI workloads require infrastructure-aware reliability, and 4.2 builds that reliability into the platform itself, so enterprise teams can build, train, and serve their own models on hardware they already own.

Fuzzball 4.2 is agent-ready infrastructure. A new MCP server lets AI agents inspect a Fuzzball environment directly and draft, submit, and monitor workflows. Writes, workload execution, and destructive actions each require explicit permission, so every agent works inside boundaries an operator sets. Separately, workflows can now reach the Fuzzball API themselves. Every job and service container receives a workflow-scoped API credential and the cluster's API address automatically, so a running job or service can submit, track, and stop additional workflows with no separate login and no long-lived shared secrets. That opens the door to more complex Fuzzball-native pipelines, and to hosted services that broker access to Fuzzball-managed compute for users who never touch the cluster directly. Taken together, an AI agent can now drive the platform, and the platform can now run an agent that does the same.

"Enterprises have moved their AI in-house faster than they have built the operations to run it. What decides whether that works is the software layer above the hardware, and Fuzzball 4.2 is that layer. Teams build, train, and serve on one platform, on infrastructure the organization governs, with their models and their data staying where they belong," said Gregory Kurtzer, CEO and founder of CIQ and founder of Rocky Linux.

Fuzzball's updated scheduling collects host and GPU health signals, assigns every node a reliability score, and incorporates that health directly into scheduling decisions. Jobs restart elsewhere when a node fault is detected, and Fuzzball automates controlled cordoning, draining, and replacement of degraded hardware, so organizations get more useful work from expensive infrastructure.

Production AI infrastructure must also safely serve more than one team. Organization-level storage isolation, default storage provisioners, and compute-policy grants allow enterprises and service providers to share large GPU environments while they maintain clear boundaries around data, infrastructure, and placement rights. Those boundaries rest on architecture rather than policy alone, because Fuzzball executes workloads unprivileged and rootless, an approach inherited from high-performance computing systems shared by thousands of users. Per-workflow accounting for compute, storage, and network egress gives operators visibility into what each workload consumed.

These capabilities change what an organization can do with accelerators it already owns. One GPU environment serves many teams at once, with enforced boundaries around each team's data, storage, and placement rights. Inference services run as persistent endpoints in the same platform as the training that produced them, and the pipelines and agents that drive that work run there too. Models, weights, and data stay on infrastructure the organization owns and governs, on premises or in any supported cloud, with the same workflow definition in both places and every action auditable.

AMD ROCm GPU support expands Fuzzball's ability to discover and schedule workloads across both NVIDIA and AMD infrastructure. As GPU vendors acquire the schedulers their customers depend on, Fuzzball remains hardware agnostic and independent of any single accelerator supplier, so enterprises buy the hardware that fits the workload without changing how they operate it.

Availability

Fuzzball 4.2 is available now. Fuzzball deploys to AWS, Google Cloud, Oracle Cloud, CoreWeave, and Azure, as well as on-premises clusters built with Warewulf, VMware, or bare metal. To get started, contact CIQ at ciq.com/products/fuzzball.

About CIQ

CIQ is the founding support and services partner for Rocky Linux and a leading provider of Enterprise Linux infrastructure. CIQ delivers commercially supported Linux offerings, high-performance computing solutions, and AI infrastructure to enterprises, government agencies, research institutions and supercomputing centers worldwide. CIQ's products include the Rocky Linux from CIQ (RLC Pro) family of operating systems, Ascender Pro for IT automation, Fuzzball job-based container orchestration, Warewulf cluster provisioning and Apptainer, the leading container system for high-performance computing. For more information, visit ciq.com.

Contact:
Stephanie Schlegal
ciq@offleashpr.com