New 'gpuc' Tool Auto-Shuts Down Expensive AI Rentals When Idle

An independent researcher named Brendan Long needed a simple way to run AI experiments across several graphics chips (GPUs, the expensive hardware that trains AI models) without wasting money. Nothing off-the-shelf did the job, so he built his own tool, called gpuc, and shared it for free on September 29, 2026. It queues up work and sends each job to whichever machine is available.

The problem it solves is money. Renting a powerful chip from a cloud service like RunPod costs roughly $2–4 per hour, and the classic mistake is renting one, finishing your work, and forgetting to shut it down — sometimes until 2 a.m., sometimes all weekend. gpuc watches the queue and shuts the rental off automatically once nothing is waiting and a timeout passes. It also removes leftover files after successful runs, so storage bills don't creep up either.

It works like a deli counter: jobs line up, run in priority order, and you can check status or watch the live output. You can hand it your own machine, a shared university server you can reach over a network, or a paid cloud rental. Importantly, it needs no administrator password, which matters because many people only have limited access to shared computers. It also plugs into Claude (the AI chatbot) so you can manage jobs by typing a command into a conversation, and into Snakemake, a tool that chains analysis steps together.

The honest limits: only one person can use each setup, RunPod is the only rental service supported so far, it doesn't handle AMD chips, and it won't run jobs inside containers or Kubernetes (the standard corporate way of packaging software). Long says he built it mainly for solo researchers who don't have a lab's budget — a reminder that a lot of AI progress happens on shoestring money, and that forgetting to flip a switch is a real cost.