Homelab & local AI
Placeholder
(Ryzen AI MAX+ 395, 64 GB)
The problem
What I built
- Platform — k3s, Traefik, ArgoCD (GitOps), Prometheus + Grafana + Loki
- Inference — llama.cpp + llama-swap on ROCm, model routing per task
- Services — n8n, ComfyUI, Immich, Synapse, AdGuard (podman quadlet)
- Isolation — agents in a KVM VM behind an egress-deny boundary