This will deploy to nodes labeled as GPU and create endpoints that anyone on your Tailnet can access, all wired up and ready to go.

As an example, once initial setup is done - you can run Ollama, Open-WebUI and automatic1111 stable diffusion image generation with:

pulumi config set ollama:enabled true
pulumi config set open-webui:enabled true
pulumi config set automatic1111:enabled true
pulumi up

The cluster works even on a single laptop node but it's a great way to use spare CPU cycles and storage on other Linux hardware you have available.

For now it's mostly to setup AI components but it's easy to add anything that can be deployed to Kubernetes (either Helm charts or just docker images).

Good engineers figure out the final requirements on first go, sure...