THE PLAN.
- Stage 1 · Now
Prove it works.
Founding homes pool their graphics cards to train our first real AI model from scratch, across homes. Founding homes volunteer spare compute: no pay yet, and no cost to you.
- Stage 2 · Next
Open the compute marketplace.
Once the network is proven and growing, companies and researchers buy AI compute from HETRA, and the money goes to the homes whose cards did the work. Paid in dollars. No tokens, no crypto.
- Stage 3
Grow it.
More homes, bigger models, and a bigger share of the AI economy going to the people who power it.
HETRA has no token, coin or crypto, and never will. If you see a “HETRA token” or anyone selling one, it isn’t us. Our only official site is lluscien.com and our only email is [email protected].
What founding homes get
- Name the model. Vote on the name of the first model you help train.
- Your name on it. Credited, if you want, on the model’s “trained by” page and in our research.
- Free use of the HETRA models you helped train.
- First in line when the marketplace opens and paid work starts.
- A say in what’s next. Vote on what we train after that.
- Founding Home #N, for good, plus early access to the HETRA app.
We’re developing the systems required to bring those machines together, distribute useful AI workloads, verify completed work, and account for each participant’s contribution.
Our research has already demonstrated distributed inference and collaborative training using ordinary consumer graphics cards under controlled, simulated residential network conditions. Read the paper ↗
The next challenge is building a real-world network that can deliver those capabilities reliably and economically.
The goal isn’t merely to connect computers. It’s to make their collective power valuable enough that companies want to buy it.
The 9,000-answer run, start to finish.
This is a real run on household graphics cards, played back at four times speed. A pool of homes answers 9,000 questions while the network does what a real one must: lets homes join, survives one going dark, catches a cheater, and pays everyone exactly for their work. Tap a moment to jump to it.
- Two homes join. The coordinator starts handing out questions, and answers start flowing.
- More homes plug in. Each new card raises the pool’s throughput, shown by the gold line.
- A cheater joins, claiming to be the fastest home but sending made-up answers. Audits catch 7 of its 8 checks, and it’s banned 1.1 seconds after joining. It forfeits its pay, and honest homes redo its 81 answers.
- A home goes dark mid-run. Its 6 unfinished questions move to other homes, and nothing is lost.
- 9,000 of 9,000 delivered. $3.60 paid out, each home’s share matching the verified work it did, to the cent.
Recorded from the Replay tab. The full run took 194 seconds. Raw logs and a verifier ship with the benchmark.
GOT A GAMING PC?
Free · 30 seconds · just your email and graphics card
Distributed AI computing. Independently owned hardware. A shared opportunity.
HETRA is an early-stage research project. Paid compute services and participant compensation are not yet available. Future earnings will depend on customer demand, hardware capabilities, operating costs, and verified work.