Your hardware earns its keep
Point a machine at the network and it starts serving a model. Stake $HERO behind it. When people use that model, the fees flow to the machines that served it, split by what each one actually proves: capacity, speed, and uptime. Better hardware, bigger share. Yield arrives as API credits on your key, and your staked $HERO comes back whenever you unstake it.
Loading…
One command joins a machine. It downloads a small model, serves its share of the experts, and reports in with your key.
pip install "mlx-expert-sniper[network] @ git+https://github.com/walter-grace/expert-sniper.git" mlx-sniper download olmoe-1b-7b -o ~/models/olmoe-stream expert-node --model-dir ~/models/olmoe-stream \ --partition 0-63 --join YOUR_API_KEY \ --advertise-url http://your-public-url:8301
The network checks your node by fetching random expert blocks and verifying them byte for byte. No benchmarks to fake. It either serves fast or it does not.
Stake weighs in at its square root, so hardware stays the main event. Principal is never at risk from a slow day; a machine that underperforms simply earns less.
Loading…