Personal GPU-rental harness — provision a vLLM pod on Vast.ai or RunPod, expose it over Tailscale, terminate when the daily budget is spent.

A pure, zero-dependency denial-of-wallet spend guard: given this window's spend, may I make one more paid action? For email sends, LLM calls, or any paid API. Runs anywhere.