> ## Documentation Index
> Fetch the complete documentation index at: https://docs.polystack.tech/llms.txt
> Use this file to discover all available pages before exploring further.

# On-Prem AI Infrastructure Platform

> Run GPU-accelerated VMs, Kubernetes clusters, and end-to-end MLOps on dedicated on-premises infrastructure with NVIDIA, Intel Gaudi, and AMD accelerators.

The **Polystack On-Prem AI Infrastructure Platform** delivers a complete, private AI stack on
your own hardware. GPU-capable virtual machines and Kubernetes clusters, a built-in operator
catalog, runtime security scanning, a ready-to-use MLOps toolchain, and accelerator-level
performance tuning are managed together from one console, across NVIDIA, Intel Gaudi, and AMD
accelerators.

<Note>
  The platform builds on [Polystack Compute](/services/compute) for GPU-capable virtual
  machines and on [Polystack Kubernetes as a Service](/services/kubernetes/index) for GPU-enabled
  container clusters.
</Note>

***

<p style={{ fontSize: '1.25rem', fontWeight: 700, marginBottom: '0.75rem' }}>Platform Capabilities</p>

<CardGroup cols={2}>
  <Card title="GPU-Capable VMs and Containers" icon="server" href="/services/ai-platform/gpu-vms-and-containers" color="#bf9667">
    GPU passthrough and GPU sharing for virtual machines, plus Kubernetes clusters with
    dedicated GPU node groups.
  </Card>

  <Card title="Operator Framework" icon="puzzle-piece" href="/services/ai-platform/operator-framework" color="#bf9667">
    Operator Lifecycle Manager and the OperatorHub.io catalog, installed by default on every
    Kubernetes cluster.
  </Card>

  <Card title="Runtime Vulnerability Scanning" icon="shield-halved" href="/services/ai-platform/vulnerability-scanning" color="#bf9667">
    StackRox runtime protection across all clusters, with Harbor and built-in Trivy for
    image and registry scanning.
  </Card>

  <Card title="MLOps Platform" icon="diagram-project" href="/services/ai-platform/mlops" color="#bf9667">
    Kubeflow, KServe, and MLflow, delivered as a one-click AI cluster template.
  </Card>

  <Card title="GPU and HPU Performance Optimization" icon="gauge-high" href="/services/ai-platform/performance-optimization" color="#bf9667">
    CPU pinning, NUMA affinity, hugepages, vendor GPU operators, GPUDirect RDMA, and
    accelerator telemetry.
  </Card>

  <Card title="Multi-Vendor Accelerator Support" icon="microchip" href="/services/ai-platform/multi-vendor-accelerators" color="#bf9667">
    NVIDIA, Intel Gaudi, and AMD accelerators, each with its own flavor, cluster template,
    and operator.
  </Card>

  <Card title="Unified Console" icon="window-maximize" href="/services/ai-platform/unified-console" color="#bf9667">
    One portal and one single sign-on for CPU and GPU workloads, VMs, and Kubernetes clusters.
  </Card>
</CardGroup>

***

<p style={{ fontSize: '1.25rem', fontWeight: 700, marginBottom: '0.75rem' }}>Platform Architecture</p>

```mermaid theme={null}
graph TD
    PORTAL[Unified Console<br/>Polystack Dashboard + Rancher<br/>Single Sign-On]
    PORTAL --> VM[GPU Virtual Machines<br/>Polystack Compute]
    PORTAL --> K8S[GPU Kubernetes Clusters<br/>Polystack K8SaaS]
    K8S --> OPS[Operator Framework<br/>OLM + OperatorHub.io]
    K8S --> SEC[Runtime Security<br/>StackRox + Harbor/Trivy]
    K8S --> ML[MLOps<br/>Kubeflow + KServe + MLflow]
    K8S --> GPUOP[Vendor GPU Operators<br/>NVIDIA / Intel Gaudi / AMD]
    VM --> HW[Accelerator Hardware<br/>NVIDIA GPU / Intel Gaudi HPU / AMD GPU]
    GPUOP --> HW
    GPUOP --> MON[Accelerator Telemetry<br/>DCGM / ROCm exporters to Prometheus + Grafana]
```

***

<p style={{ fontSize: '1.25rem', fontWeight: 700, marginBottom: '0.75rem' }}>How the Platform Fits Together</p>

<Steps>
  <Step title="Accelerators are exposed to workloads" icon="microchip">
    Polystack Compute presents NVIDIA, Intel Gaudi, and AMD accelerators to instances through
    PCI passthrough, NVIDIA vGPU, or AMD SR-IOV, using a dedicated flavor per vendor.
  </Step>

  <Step title="Kubernetes clusters consume GPU flavors" icon="boxes-stacked">
    Polystack K8SaaS provisions clusters with GPU node groups on those flavors, and a
    vendor-specific operator runs on each node group.
  </Step>

  <Step title="Every cluster ships with platform services" icon="layer-group">
    Operator Lifecycle Manager, the OperatorHub.io catalog, and StackRox Sensor and Collector
    are deployed to every cluster automatically.
  </Step>

  <Step title="AI teams start from a ready-made AI cluster" icon="diagram-project">
    The AI cluster template adds the GPU Operator, Kubeflow, MLflow, and Harbor in a single
    deployment.
  </Step>

  <Step title="Everything is managed from one portal" icon="window-maximize">
    The Polystack Dashboard and Rancher sit behind one single sign-on and are presented as a
    single unified portal.
  </Step>
</Steps>

***

<p style={{ fontSize: '1.25rem', fontWeight: 700, marginBottom: '0.75rem' }}>Related Services</p>

<CardGroup cols={2}>
  <Card title="Kubernetes as a Service" icon="boxes-stacked" href="/services/kubernetes/index" color="#bf9667">
    Cluster templates, node groups, and lifecycle management for the Kubernetes clusters the
    platform runs on.
  </Card>

  <Card title="Compute" icon="server" href="/services/compute" color="#bf9667">
    Instances, flavors, and hypervisor configuration behind GPU-capable virtual machines.
  </Card>

  <Card title="Monitoring" icon="chart-line" href="/services/monitoring/index" color="#bf9667">
    Prometheus and Grafana dashboards that receive accelerator telemetry.
  </Card>

  <Card title="Identity and Access" icon="user-shield" href="/services/identity/index" color="#bf9667">
    Projects, users, and roles that back the platform's single sign-on.
  </Card>
</CardGroup>