Built a self-healing server agent that sends inference requests from an i3 server to my laptop's GPU over Tailscale my homelab kept crashing at the worst possible times. container dies at 3am, disk fills up while i'm at work, some random process pegs the cpu and i find out hours later. got sick of it and built a thing. the setup is dumb but it works: \- old i3 with 8gb ram and a 1tb hdd — my "local cloud". runs docker, stores my stuff. this is the machine that needs babysitting. \- old i5 with 8gb ram — runs n8n in docker, exposed via cloudflare tunnel for my workflows. \- my laptop, rtx 3050 + 24gb ram — the only thing in my house that can actually run an llm. the problem: i wanted the i3 server to make smart decisions about itself, but it obviously can't run a model. and the laptop isn't always on the same wifi. fix: tailscale. the server just talks to the laptop over the mesh vpn, encrypted, no port forwarding, no public anything. firewall only lets traffic through on the tailscale adapter.…