GitHub
Run Miner
Testnet live
GPU miners contribute compute. TEE-attested validators score gradient quality cryptographically. Sui settles rewards on-chain in real time.
No application. No whitelist. Apache 2.0 at TGE.
Network status
ALREADY RUNNING
The validator is live. The contracts are deployed. The scoring loop has been running on testnet since March.
When you point a miner at the network, you are joining something that already works. You train, the enclave scores, the contract pays. That is the loop.
HARDWARE VERIFICATION
The scoring binary runs inside an AWS Nitro Enclave. Its PCR values, the cryptographic fingerprint of the exact code in memory, are registered on Sui the moment the enclave starts.
If anyone modifies the scoring logic, the PCR changes, the Ed25519 signature fails on-chain verification, and the score is rejected. It is not policy. It is how the math works.
Nobody can forge a score. Not a validator, not a third party, not us.
TOKEN ECONOMICS
21M total supply. Hard cap, fixed on-chain. 70% to network participants, earned per window through verifiable contribution. 30% pre-minted at TGE for treasury, liquidity, team, investors, and airdrop.
ALLOCATION · 21M HARD CAP
KEY PARAMETERS
VALIDATOR LOOP
EMISSION SCHEDULE
TRACTION
MARKET CONTEXT
On March 10 2026, 70+ commodity GPUs completed Covenant-72B — a 72B parameter model trained on 1.1 trillion tokens without a single data center. The infrastructure layer that makes distributed training trustworthy at scale does not yet exist.
The window for the first TEE-verified compute market on Sui is open. Every existing decentralized training network — without exception — relies on social trust for scoring. VRAM replaces that with cryptographic proof.
ROADMAP
TEAM
12+ years across AI, HFT infrastructure, and blockchain. Cross-chain migration at PAID Network. HFT systems at Broadridge. Blockchain engineering at BNP Paribas.
10+ years blockchain development with deep Rust and Sui Move expertise. Primary architect of the Rust workspace and the deployed Move modules.
Ecosystem partnerships, community growth, and go-to-market strategy.
GET STARTED
No application. No whitelist. Any NVIDIA GPU. Apache 2.0.
Trains on your GPU. Earns $VRAM proportional to loss improvement. Rewards are immediate — no vesting for miners.
Requires Nautilus TEE endpoint (AWS Nitro) and a valid Validator Ticket. Earn $VRAM per window proportional to contribution.
TRY LOCALLY FIRST
cargo run --bin vram-local-demo
5 windows · 2 miners · 2 validators · no network required
NETWORK OBJECT IDs
FAQ
Any NVIDIA GPU with 8GB+ VRAM. If your machine can run a modern game, it can mine. There is no minimum stake, no application, and no approval process.
Yes. Apache 2.0, fully open source at TGE. Trains a model slice and uploads compressed gradients. No admin access, no kernel modules.
No. Your storage credentials are encrypted on-chain via Sui Seal IBE. Validators forward gradients to the Nautilus enclave but cannot access or exfiltrate what's inside — the enclave processes them in isolation. The only entity that can decrypt your credentials is the enclave itself. Not the validator operating it, not us, not anyone with network access. Your weights stay yours.
Testnet is live now. Mainnet launches after the Move contract external audit completes. Track progress on Discord and X.
Every window, your gradient is scored against all other miners by the TEE. The better your loss improvement relative to the field, the higher your share. Scores are computed using OpenSkill Plackett-Luce rankings — the same method used in competitive gaming matchmaking. The contract distributes automatically. No human in the loop.
Three reasons. Move's object model makes on-chain credential management with Seal IBE practical. Sui's parallel execution engine processes reward distribution in under a second — every window, not just at settlement. And Nautilus TEE integration is a first-class primitive in the Sui ecosystem, not a workaround.