AI Agent

Overview

The Kubermatic Developer Platform AI Agent is a specialized assistant that helps users generate Kubernetes resource YAML files through natural language within KDP workspaces. It converts requests in natural language into properly formatted Kubernetes manifests, eliminating the need to manually write lengthy YAML files from scratch.

Prerequisites

Before installing the AI Agent, ensure you have:

• A running KDP installation on your Kubernetes cluster
• OpenAI API key for the language model capabilities
• OIDC provider configured (same one used by KDP)

Installation

The AI Agent is deployed using Helm. Follow these steps to install it:

Prepare the Configuration

Create a ai-agent.values.yaml file with your specific configuration:

aiAgent:
  imagePullSecret: |
    {
      "auths": {
        "quay.io": {
          "auth": "<PULL_CREDENTIALS>",
          "email": ""
        }
      }
    }


  config:
    oidc:
        clientID: kdp-kubelogin
        clientSecret: <OIDC_CLIENT_SECRET>
        issuerURL: https://login.<DOMAIN>
    kubernetes_api_url: https://api.<DOMAIN>
    openai_api_key: "<OPEN_API_KEY>"  # OpenAI API key for the language model

  ingress:
    create: true
    host: <DOMAIN> # Use same domain as the frontend to avoid CORS errors
    prefix: /ai-agent(/|$)(.*)
    certIssuer:
      kind: ClusterIssuer
      name: letsencrypt-prod

Before deploying the KDP dashboard, you need to replace the following placeholder variables in the ai-agent.values.yaml file with your own values:

• <OPEN_API_KEY>
• <DOMAIN>
• <OIDC_CLIENT_SECRET>

The <OIDC_CLIENT_SECRET> placeholder must be replaced with the value set in Dex and configured in the dex.values.yaml file.

Install with Helm

Now that all placeholders are replaced, deploy the KDP AI Agent Helm chart. To log into the Helm registry, use your email address as the username and the license key you received as the password.

helm registry login quay.io
helm upgrade --install kdp-ai-agent \
    oci://quay.io/kubermatic/helm-charts/developer-platform-ai-agent \
    --version=0.9.0 \
    --create-namespace \
    --namespace=kdp-system \
    --values=ai-agent.values.yaml

Configure the Dashboard

To make the AI Agent accessible from the KDP Dashboard, you need to update the values.yaml file for your dashboard deployment. Assuming you followed the quickstart, this file would be kdp-dashboard.values.yaml.

You will need to edit it to activate the AI Agent feature and set the backend url.

dashboard:
  config:
    features:
      aiAgent:
        enabled: true
        generatorURL: <DOMAIN>/ai-agent/ # same domain as the host

You’ll need to replace <DOMAIN>.

Then after this update the release of your kdp dashboard. If you followed the Quickstart it will be called kdp-dashboard in the kdp-system namespace, so the command would look like this to first login and then update:

$ helm registry login quay.io
$ helm upgrade --install kdp-dashboard \
    oci://quay.io/kubermatic/helm-charts/developer-platform-dashboard \
    --version=0.9.0 \
    --create-namespace \
    --namespace=kdp-system \
    --values=kdp-dashboard.values.yaml

After this you will need to delete the pod for the dashboard manually for it to be redeployed and pick up the new values. You can find them by the label app.kubernetes.io/name: kdp-dashboard and delete with.

kubectl delete pods -l app.kubernetes.io/name=kdp-dashboard -n kdp-system

Verify the Installation

Once the pods start, you can use the AI Agent in the frontend.

A purple button should be visible in the form to create a new service object within a workspace.

Then, once clicked, a text field will be visible were you can describe how you want your resource to be.

Here is an example after writing a prompt and clicking on Generate:

After a few seconds you should get the result:

You can then edit and modify if you like, from the form or directly in the YAML.

You also do not have to worry about getting a wrong schema since it is getting validated in the backend. You can be sure there are no hallucinated fields nor missing required fields.

Note: Please be sure to check the values and the YAML in general before submitting. AI can make mistakes.