v0.3.0

• SD-Next (https://vladmandic.github.io/sdnext-docs/) as alternative to automatic1111 stable diffusion
• InvokeAI community edition (https://www.invoke.com/)
• updates to KubeAI and vLLM
• Application builder class to make it easier to define applications
• support for adding Grafana dashboards
• allow overriding storage class used by each app
• preferredNodeLabel to deploy to subset of nodes or specific node
• requiredNodeLabel to require labeled nodes (replaces zone setting)
• cpu/memory limits per namespace, application and container
• version now optional for Helm charts, use latest by default
• faster recovery when node down (default toleration reduced from 5 minutes to 10 seconds)
• clean up application labels, use standard app.kubernetes.io labels
• support for liveness, readiness and startup probes
• use Recreate instead of RollingUpdate by default for apps with storage
• documentation updates (moved module docs to subfolders, troubleshooting section, SSH, disable suspend on lid close)

v0.2.0

AI module release:

• NVidia GPU Operator
• Ollama
• Open-WebUI
• automatic1111

Beszel (https://beszel.dev/) monitoring component.

Longhorn as default provisioner.

Move from oranglelab:<app> to <app>:enabled setting.
<app>:storageOnly to disable application but keep storage.

Read-only k8s access with tailscale configure kubeconfig k8s

Pulumi IDs updated to group by module.