As artificial intelligence (AI) moves deeper into CRM and enterprise systems, it is opening a performance pathway that brings high-performance computing (HPC) capabilities to more industries.
CIQ, the founding support and services partner of Rocky Linux and developer of high-performance software infrastructure, in January announced Service Endpoints, a new capability for its Fuzzball platform.
Fuzzball bridges the gap between traditional, complex HPC environments and cloud-native, containerized architectures, enabling flexible deployment across on-premises, edge, and hybrid cloud environments.
Training often happens on one platform and inference on another — a process slowed by manual handoffs and unbudgeted technical debt. CIQ’s latest update to Fuzzball eliminates that friction, turning that divide into a single, portable Linux workflow that actually respects your uptime.
This new approach simplifies complex HPC workflows into manageable, automated tasks. It enables high-performance analytics, AI model training, and inference for CRM data without requiring deep infrastructure expertise.
According to Jonathon Anderson, Fuzzball Product Lead at CIQ, organizations want to use AI with their proprietary data in their own controlled environments without sending it to external platforms.
"Fuzzball Service Endpoints treats the entire AI stack as a composable unified workflow that can be iterated on and refactored to meet specific requirements. This gives you complete control over your AI experience, whether you’re fine-tuning a model or orchestrating a suite of coordinating agents," he told CRM Buyer.
More Than Just a Shiny GUI Skin
At its core, Fuzzball’s Service Endpoints capability addresses a fundamental architectural friction: the wall between batch processing and persistent services. Traditionally, a Linux cluster is well-suited to processing a massive training dataset (batch). Still, it struggles to keep an inference API or persistent service alive without a messy web of manual networking and resource carving.
CIQ’s approach unifies these disparate worlds into a single, portable workflow definition. By treating the entire AI lifecycle as a composable unit — ingestion, training, and the persistent endpoint — Fuzzball eliminates the manual handoff that typically leads to brittle deployment pipelines.
Anderson explained that for enterprise Linux users, this means the same YAML-defined workflow that fine-tunes a model on a local workstation can be deployed to a massive hybrid cloud cluster and run without modification. The result maintains bare-metal performance while providing the high-level service flexibility that modern CRM applications demand.
Fuzzball Substrate is the custom runtime making that happen. Unlike standard Kubernetes, which often has a "sidecar tax" (networking overhead), Fuzzball lets containers talk directly to the hardware. That makes the CRM benefits much more believable to a skeptical sysadmin.
Fuzzball Exceeds a Standard Cloud CRM Setup
In a typical cloud-native (Kubernetes) environment, a virtualization tax — often 5% to 25% — is introduced by the hypervisor and complex Container Network Interface (CNI) plugins. For real-time CRM tasks — like an AI agent scanning a million customer records to provide a live recommendation — that latency adds up.
Fuzzball Substrate is built around three performance-focused enhancements:
Direct Hardware Access: Acts as a specialized execution agent that bypasses the standard Kubernetes networking overhead. That direct hardware access allows the containerized AI model to communicate directly with the CPU, GPU, and high-speed storage.
Custom Container Runtime: Leverages the Singularity Image Format (SIF), the gold standard in HPC for performance and security on shared systems.
Eliminating Handoff: Orchestrate handles the scheduling. The Service Endpoint runs in the same optimized environment in which it was trained. No data movement. No reformatting. No performance loss.
From Development to Deployment
Developers often favor local workstations for their ease of use and accessibility. Fuzzball simplifies the development of reproducible workflows on centralized resources.
Since all Fuzzball jobs and services are containerized, users can run a container on a local workstation or execute individual workflow components locally during testing and development.
"Not having to build separate artifacts for different environments means there’s less to manage, and the developer, testing, and production experiences are more likely to coincide," he offered.
Competing With Proprietary AI Clouds
A major benefit is that it allows companies to keep their data strictly on-premises. Fuzzball makes it easy to deploy platforms on your own infrastructure, whether open-source or proprietary.
"We’ve been particularly focused on combining open-source AI platforms with Fuzzball," said Anderson. "AI is a fast-developing field, of course, and larger models may be available through third-party AI services," he admitted.
Using that type of access often limits control over your data. Self-hosting your AI infrastructure keeps your data under your control, Anderson noted.
"The self-hosted models are getting better every day, just like the models available through popular third-party services," he added.
Acts as Its Own OS for Bare-Metal Performance
In an enterprise Linux environment, Fuzzball acts as the operating system for AI agents to communicate and execute tasks without human intervention. The platform acts as a resource manager and application scheduler, enabling workflows to run any application on any infrastructure.
"It is fundamentally API-driven and, in the future, we anticipate AI agents running within the Fuzzball system to be able to interact not only with the outside world through the growing ecosystem of MCP-enabled systems, but reflectively with Fuzzball itself," he hinted about future development plans.
Reinforcing Linux’s Role in Enterprise AI
CIQ is at the forefront of making enterprise Linux more essential for CRM and beyond. As the founding partner of Rocky Linux, CIQ ensures that Fuzzball fully leverages the Red Hat Enterprise Linux (RHEL)-compatible ecosystem.
"We build and test Fuzzball on Rocky Linux and benefit from the long-term stability of the Enterprise Linux platform and the accessibility and collaboration of the Rocky Linux community," Anderson said.
In the future, he expects to further benefit from performance-optimized and hardened variants of Rocky Linux. But Fuzzball’s fully containerized workflows mean users can build jobs on nearly any modern Linux distribution.
Loading ...