Homelab
Proxmox cluster with RTX 3090 passthrough that runs my self-hosted LLM stack.
A small but realistic setup that mirrors common production patterns: a hardened public edge, private compute, and controlled exposure of services.
Layout
- Public edge: a VPS (
netwatch) is the only internet-facing entry point. It terminates TLS and routes traffic over WireGuard. - Private core: a Proxmox cluster runs VMs and containers, separated by purpose.
- GPU node: an RTX 3090 with PCIe passthrough serves models through vLLM.
- Private access: Tailscale for admin access. Management planes are never exposed.
What runs on it
vLLM, Open WebUI, LiteLLM, Langfuse, SearXNG and n8n. A Kubernetes cluster on KVM nodes is the operations playground: ingress, certificates, storage, upgrades and rollbacks.
Why
It forces good habits: DNS and TLS automation, least-privilege exposure, and a clean line between the internet edge and the compute layer.
[TODO: architecture diagram]