Remove KubeOn

1. Remove the agent from each cluster

helm uninstall kubeon-agent --namespace kubeonkubectl delete namespace kubeon

• Flux or Argo CD: delete the HelmRelease or Application from your repository and let the controller prune it.
• Terraform: remove the agent's helm_release resource and apply.

This removes the controller too, if you turned it on. Then delete its separate cloud automation role in your account.

kubectl get namespace kubeon                              # NotFoundkubectl get clusterrole,clusterrolebinding kubeon-agent-reader   # NotFound

Then revoke each cluster's token in Settings > Integrations, so a copy left anywhere can no longer send data.

2. Revoke cloud access

# Billing role, deployed with CloudFormationaws cloudformation delete-stack --stack-name kubeon-billing-accessaws cloudformation wait stack-delete-complete --stack-name kubeon-billing-access

The agent used no cloud credentials, so there is nothing to remove for it. Your billing exports, the CUR, Glue and Athena tables, BigQuery datasets and Azure exports were created by you and are left as they are.

3. Clear snapshots

Snapshots are in your organization's bucket in KubeOn, not in your accounts. They are deleted with the workspace, as agreed in your evaluation or contract; to have them deleted sooner, email privacy@unitvane.com.

4. Close the workspace

Ask us to close your workspace, and your UnitVane organization if you use it. Your data is deleted or kept as agreed in your evaluation or contract. Export reports, or rotate and then revoke your Bring your own BI credentials, before you do.

5. Remove the UnitVane AI gateway

1. Point each application back at its provider's endpoint, or another route, and deploy. Revoking keys first would make those requests fail.
2. Revoke the gateway's team and application keys.
3. Remove the provider credentials created for the gateway, such as the Bedrock role or the Azure OpenAI workload identity, and the read-only billing access for AI tokenomics.

6. Remove the KubeOn GPU optimizer

1. Set every GPU policy to Recommend, so the optimizer stops acting.
2. Drain any burst nodes back home, or let the burst policy return them, before you remove its cloud credentials.
3. helm uninstall kubeon-gpu-optimizer --namespace kubeon-gpu, then delete the gpu.kubeon.io CRDs if you no longer need the policies.
4. Remove the cloud credentials you granted for burst locations.

What stays