A public neural network · devnet live & training

An AI no one owns.
Everyone builds it.

One living model, trained on spare GPUs like yours, recorded on a public chain the way a ledger records money. And it pays you for the work.

How it works

$ $ $ $ $ $ $ YOUR GPU A GAMING PC AN OLD LAPTOP PEOPLE APPS AGENTS YOUR DATA named · credited · paid THE MODEL one public brain, on-chain your work $ back to you $ fees in answers back teaching royalties

  1. You train it

    Spare machines like yours teach one shared model, in rounds sized to what your card can do.

  2. Your data teaches it

    Every update names the data it learned from, and the chain keeps that record forever. Your name stays on your work.

  3. The world pays to use it

    Every answer the model gives costs a small fee, paid by whoever asked.

  4. You get paid

    Fees run back down the same lines: wages for your GPU’s work, royalties for your data.

Today’s best AI belongs to a handful of companies. This one belongs to you: the model lives on the chain itself, trained by machines like yours, rebuilt bit for bit on your own desk. The thing being mined is intelligence.

The deal

The model doesn’t live on our servers. It lives on the public chain, where you can read it, check it, and rebuild it on your own machine. No gatekeeper, no API key, no one who can ever cut you off.

Point your graphics card at the network and it trains the live model. Your gaming PC, your workstation, even your Apple laptop. The client measures your speed and sizes every round to what your hardware can actually finish. No minimum rig, no permission slip.

Train, and you earn. Every block splits its reward between the people who improved the model and the owners of the data it learned from. Every payout is public math you can check yourself.

No other network does this: when enough training is flowing in, the chain votes to make its own model bigger. No committee, no admin, no new download. The brain you’re paid to build grows because you fed it.

Get paid by users, not hype.

Most networks pay you in freshly printed tokens and hope they’re worth something later. Sestrian charges a fee for every answer the model serves and splits it between the people whose GPUs and data built it. Real usage funds rewards, not a money printer.

Today this is the devnet: the proving ground. Devnet coins exist to test the economics, they are not traded anywhere, and the chain can still reset. Mainnet comes once the protocol has earned it.

The goal: the crossover. The first block where users pay in more than the printer mints. No network has ever reached it.

Four steps. The network does the rest.

  1. Bring your GPU

    One command builds the node, creates your wallet, and rebuilds the starting model on your own machine, so you never have to trust a download.

  2. Train a round

    Your machine improves the current model on real data, entirely on your own hardware, in a round sized to what your GPU can finish in time.

  3. Sign your work

    Your machine signs its improvement, stamps it with the data it learned from, and sends it out. Your name is on your work.

  4. Earn

    The network tests every update on questions it held back, keeps what genuinely helps, and pays you in proportion to how much yours did.

No GPU? Bring data.

Most models are trained on data nobody asked for and nobody paid for. Sestrian pays its sources. Stake data you have the rights to by putting a deposit behind it, and it earns: a share of every block it helps train, plus royalties when the model uses what it taught. Every claim is signed and public. If your data matters years from now, it earns then. Usage is the clock, not age.

Three ways to participate
RoleYou doYou earn
Minerun one installer and let your GPU train the modela reward every time your work makes the cut
Servehost the model and answer paid requeststhe fee on every request, straight to your wallet
Stake databack data you own with a deposita share of every block it trains, plus royalties when it’s used

Two rival GPUs. One mind.

Whatever machine you bring, the network agrees with it, exactly. Everything on this chain, from the model to the payouts, can be recomputed on your own hardware and comes out identical to the last bit. You never have to trust another machine. Or us.

devnet · reading the chain

The chain, right now.

Not a screenshot. These are the numbers the network is agreeing on as you read them. Check them against any node you like.

height
–
supply
–SES
peers
–
miners
–
model size
–
times it has grown
–
head      —
model     —

Connecting…

Phase 1 · open devnet, live now

Join the
devnet ↗
terminalchecks your GPU before you commit hours
$ git clone https://github.com/sestrian/sestrian && cd sestrian
$ scripts/install.sh --mine   # or plain install.sh to watch & sync

Verify, don’t trust: your node rebuilds the starting model itself and checks every byte before it joins.