Train AI on hardware people already own

Graphics cards sit idle in bedrooms while people with data cannot afford to train anything. This puts the two together.

I have data

Train a model

Upload a dataset and send it to a machine with a spare GPU. You get the trained model back, checked against data the machine never saw.

No graphics card needed. Any computer will do — the training happens on somebody else's.

Sign in with GitHub

Already have one?

Reads your GitHub username, nothing else. Your key is never sent to GitHub, and the coordinator stores only a one-way digest of it.

I have a GPU

Lend your graphics card

Run a small agent that takes training jobs when your machine is idle. It stops on its own if the card gets too hot, and you approve each job unless you say otherwise.

Needs an NVIDIA GPU on Windows with WSL 2, or Linux.

Set up your machine
Working…

How a job runs

  1. You send a dataset

    A spreadsheet, and which column you want predicted. Part of it is held back before anything is sent out.

  2. An idle GPU trains it

    The job goes to a machine somebody offered. Its owner can approve each one, and it stops on its own if the card gets too hot.

  3. The result is checked

    The coordinator scores the returned model itself, on the rows that were held back — so the grade does not come from the machine being graded.

Already set up? Your browser remembers — pick a door and you will be taken straight through.