Your GPU. One link. The world.

Share your GPU with anyone, anywhere. No accounts, no cloud, P2P.

Get RunSnack Need or have a GPU? Book on Discord

Instant

Paste a link, get a terminal.

Sandboxed

Docker isolation, no host access.

Global

Works in any browser, any device.

Free

No fees, P2P, you set the price.

How it works

Before you start: RunSnack runs in Docker. If you don't have it, install Docker Desktop first (free) and make sure it's running — whale icon in system tray.

  1. 1

    Download the installer for your OS

    Windows (.ps1) Linux & macOS (.sh)
  2. 2

    For Windows: Run it in PowerShell:

      powershell -ExecutionPolicy Bypass -File snackup.ps1

    For Linux & macOS: Run it in Terminal:

      chmod +x snackup.sh && ./snackup.sh

    RunSnack quick config after execution:

      Answer: CPU cores, RAM, optional schedule — press Enter for defaults. First run gets a ~4 GB image (takes minutes — normal). After that, sessions start instantly.
  3. 3

    Share the link

      The script prints a link once everything's ready. Anyone who opens it gets an instant GPU terminal in their browser.

Why not just rent a GPU?

You can. Vast.ai, Salad, Akash and RunPod all rent you a stranger's GPU by the hour, and if that's what you need, they're good at it. RunSnack solves a different problem: getting someone specific onto your hardware, right now. A student on the lab's GPU. A maintainer on your Jetson to reproduce a bug. A collaborator on the robot in the next building. No account for them to create, no listing to publish, no marketplace to wait for — just a link. That's why it works the moment you install it, with a network of one.
Comparison table of RunSnack against Vast.ai, TensorDock, SaladCloud, RunPod Community, Akash, io.net, Nosana and CloudJetson across core model, signup, fees, time to first shell, access method, hardware breadth and custody.

What RunSnack deliberately doesn't do

No escrow, no payments, no SLA. If money changes hands, that's between you and the other person and we're not involved. Sessions are sandboxed in Docker, but Docker is a sandbox, not a hypervisor — share with people you'd trust with an SSH key, and read the before you share with anyone else.

Legal & Privacy

Contact the author Manuel Herrador: github.com/PacifAIst

FAQ

Is it safe?
Sessions run inside a sandboxed Docker container. Whoever you share a link with gets a terminal inside that container, not your host machine. The container is read-only, privilege-dropped, and wiped after each session.
How is this different from Vast.ai, Salad or RunPod?
Those are marketplaces: you rent a stranger's GPU by the hour, through an account, with the platform in the middle. RunSnack is a tool — you hand your own hardware to someone you choose, in one link, with nobody in between. It also runs on anything with Docker: a Jetson, an old card, a laptop, not just what a marketplace happens to stock.
How much does it cost?
RunSnack is FREE. You negotiate price directly with the renter — there's no platform fee & no middleman.
What GPUs work?
Any NVIDIA GPU with Docker and the NVIDIA Container Toolkit installed, e.g., from a Jetson Nano to RTX 5080 or an B200. Technically, any machine works — even CPU-only. But RunSnack is built for GPUs.
What commands can I use inside a session?
  1. Upload files: drag and drop any file onto the terminal window — it lands in /home/snackuser/uploads/.
  2. Download files: type snack-download /path/to/file and your browser will download it.
  3. Close the session (renter): just close the browser tab.
  4. Stop sharing (host): run docker stop snack1 or use the TUI with docker exec -it snack1 snackctl.
How do I book a RunSnack and get paid?
Arrange it directly with the renter on our Discord. PayPal, crypto, whatever works. Or skip payment entirely: share with a friend for free, or use it to access your own machine remotely.