Your Code, Someone Else's Compute

A launchpad for open-source compute. Verified maintainers raise USDC on Solana for their repo's GPUs, and the people who fund it get pool shares backed by that compute, not a thank-you note.

Explore raises
Fund a repo. Own a share of its compute.
The launchpad

A compute market for every repo, funded in rounds.

1

Launch

A GitHub-verified maintainer opens a round: a target in USDC, a deadline, and the repo's one share token, the same one direct funding mints.

2

Contribute

Anyone puts in USDC. Everyone in a round gets the same price: the pool's value per share when the round fills, one share per USDC for a fresh pool.

3

Fund or refund

Hit the target and every cent goes into the repo's GPU pool. Miss it by the deadline and every contributor takes their USDC back, in full.

4

Use it

Shares are access: each one buys daily requests to the repo's model through RepoGPU. When the pool runs low, the maintainer opens the next round.

Launch a raise
How it works

Paste a repo, then let its users pay for the GPU.

1

Fund

Paste any public GitHub repo and send USDC. The pool and its share token open on the first deposit, and anyone can add to it after that.

2

Verify

The repo's admin signs in with GitHub. Once GitHub confirms they are an admin, the wallet they pick is recorded on Solana as the maintainer.

3

Run

The maintainer picks a GPU and the most they will pay per hour, then switches compute on. The meter starts that second.

4

Bill

The operator bills the time compute was on, at or under that rate, with a receipt hash for every job. No second is billed twice.

Compute goals

Support you can measure, in GPU hours.

Maintainers set the compute their repo needs. The community funds the target, progress stays public on chain, and compute starts when the maintainer is ready.

10GPU hours

About half a day of one GPU. Enough to run evals on a release or keep a demo up over a launch.

100GPU hours

Four days of one GPU around the clock. A small fine-tune, or a public endpoint through a busy week.

1,000GPU hours

Six weeks of one GPU, or a week across six. The scale of a real training run.

The rules

Nobody bills more than the maintainer agreed to.

Rate capped on chain

A charge must satisfy amount × 3600 ≤ rate × seconds. The program rejects anything above the maintainer's hourly rate.

Every second billed once

Billing moves a clock forward. The same second cannot be charged twice, and a second that has not happened yet cannot be charged at all.

The maintainer holds the switch

Only the GitHub-verified admin can switch compute on. They can switch it off at any time, and so can the operator when a pool runs dry.

Your share comes back

Funders hold the repo's share token. While compute is off and the last run is settled, burn shares and your slice of what is left comes straight back to your wallet.

No rate changes mid-run

The GPU and the hourly rate can only change while compute is off, so every run is billed at the terms it started under.

No key moves a balance

The owner can rotate the operator. There is no instruction that lets anyone take a pool's USDC outside a capped charge or a holder's own withdrawal.

Compute is paid on Solana.

Pools hold USDC on Solana, priced in the same dollars GPU providers bill in. Charges go to the operator's treasury to pay the GPU provider, and every one of them is logged in its transaction with a receipt hash anyone can read.

USDC

What pools hold

1

Share token per repo

1s

Billing granularity

GitHub

Admin check for maintainers

Open source runs on borrowed GPUs now.

The people who use a model can keep its endpoint up. Fund the repos you rely on and watch the GPU hours add up on chain.

Explore pools