How to manage Red Hat OpenShift AI dependencies with Kustomize and Argo CD

March 13, 2026
Davide Bianchi Andrea Tarocchi
Related topics:
Artificial intelligenceGitOpsKubernetes
Related products:
Red Hat AIRed Hat OpenShift AI

Red Hat OpenShift AI provides a platform for data science and AI workloads, but managing its external dependencies can become complex, slowing down setup and maintenance.

To streamline this workflow, we created the odh-gitops repository. This repository provides a declarative, version-controlled, GitOps-ready template for deploying OpenShift AI dependencies on OpenShift. It includes predefined configurations to help you manage the entire dependency lifecycle.

The core idea is simple: every operator dependency is expressed as Kustomize manifests. A single oc apply -k . command or an Argo CD Application installs everything.

Why Kustomize?

Kustomize is an open source configuration management tool that lets you customize YAML manifests for different environments. It is now included in both the kubectl and oc command line interfaces (CLIs).

The odh-gitops repository uses Kustomize to manage the OpenShift AI dependencies. This GitOps approach lets you customize configurations and helps prevent conflicts by skipping resources already in your cluster.

A note on Helm: A Helm chart alternative is also being developed in the same repository (under the chart/ directory) for teams that prefer a values-driven approach. We expect to release that option in the coming months. The Kustomize-based workflow described in this post is the primary and recommended method today.

Getting started

You can deploy this GitOps configuration using two primary methods: the OpenShift CLI (oc) or Argo CD for automated synchronization.

Prerequisites:

  • A Red Hat OpenShift cluster (version 4.19 or later)
  • oc or kubectl CLI installed
  • Cluster admin permissions

Install using the CLI

The CLI method offers manual control over the installation process.

  1. Clone the repository: Start by cloning the odh-gitops repository to your local machine.
  2. Review and customize: By default, almost all the necessary dependencies for the OpenShift AI components are enabled. Before applying, you can review the configurations and make any changes. More information in the Customize the installation section.

  3. Install the dependencies: Use the oc apply -k command with the path to your Kustomize directory. This processes the Kustomize files and applies the resulting manifests, installing all required dependencies.

    oc apply -k dependencies

  4. Once the operators are Ready, install the custom resources to configure the dependencies:

    oc apply -k configurations

And that's it! All OpenShift AI dependencies are now installed and configured in the cluster.

Important note on default dependencies

Not all dependencies are installed by default, as some require manual configuration. Always refer to the README for full instructions. Before applying the configuration to your cluster, you can validate the resources that will be installed by running the following command: oc kustomize build <path-to-kustomize-dir>

Install a specific dependency

You can also choose to install only one operator and its specific dependencies. For example, if you only require the Kueue component, which also requires the cert-manager dependency. You can install both using:

oc apply -k dependencies/operators/kueue-operator

Once the Red Hat build of Kueue operator is Ready, you can install its configuration:

oc apply -k configurations/kueue-operator

Video demo

Install using Argo CD

The Argo CD method provides a GitOps approach with automated synchronization and consistency monitoring.

Prerequisites:

  • Argo CD must be installed and configured to deploy resources on your cluster.
  • Your Argo CD instance needs to be configured with permissions to handle all actions on kueues.kueue.openshift.io and kuadrants.kuadrant.io groups.
  • Fork the odh-gitops repository and apply your customizations.

Once the prerequisites are met, you can install all the dependencies by creating an Argo CD application. Since operators install Custom Resource Definitions (CRDs) asynchronously via Operator Lifecycle Manager (OLM), you must enable the SkipDryRunOnMissingResource sync policy.

A sample Application manifest is provided below. Important: Before using this manifest, you must modify the repoURL and targetRevision fields to match your own repository fork and the desired version.

apiVersion: argoproj.io/v1alpha1
kind: Application
metadata:
  name: rhoai-dependencies
  namespace: openshift-gitops
spec:
  project: default
  source:
    repoURL: https://github.com/opendatahub-io/odh-gitops.git # Match your fork
    targetRevision: main  # Match your target revision
    path: .
  destination:
    server: https://kubernetes.default.svc
  syncPolicy:
    automated:
      prune: true
      selfHeal: true
    syncOptions:
      - SkipDryRunOnMissingResource=true

To install only selected operators, you can modify dependencies/operators/kustomization.yaml file in your fork to comment out unwanted dependencies. See the Customize the installation section for more information.

Video demo

Customize the installation

The main customization point is the dependencies/operators/kustomization.yaml file. It lists every operator as a Kustomize component reference:

apiVersion: kustomize.config.k8s.io/v1beta1
kind: Kustomization
components:
  - ../../components/operators/cert-manager
  - ../../components/operators/kueue-operator
  - ../../components/operators/cluster-observability-operator
  - ../../components/operators/opentelemetry-product
  - ../../components/operators/leader-worker-set
  - ../../components/operators/job-set-operator
  - ../../components/operators/tempo-product
  - ../../components/operators/openshift-custom-metrics-autoscaler
  - ../../components/operators/rhcl-operator
  # Uncomment to enable optional operators:
  # - ../../components/operators/nfd
  # - ../../components/operators/nvidia-gpu-operator
  # - ../../components/operators/maria-db

Comment out a line to skip an operator. Uncomment to add an optional one. The same applies to configurations/kustomization.yaml for post-install Custom Resources.

For deeper customizations (changing a subscription channel, pinning a CSV version, adjusting a CR), you can either edit the manifests directly in your fork or use standard Kustomize patches in an overlay.

Conclusion

Using the odh-gitops repository and Kustomize simplifies how you manage Red Hat OpenShift AI dependencies. This method provides flexibility for customization and straightforward processes for both initial deployment and ongoing updates, whether you choose the manual control of the CLI or the automation of Argo CD.

To continue simplifying this process, we are actively working on a Helm chart to provide an alternative installation path for Red Hat OpenShift AI users who prefer a native Helm experience.