Hanzo Network
hanzo.network is the fabric that hanzod nodes form. Put a machine on it and it serves models to whoever the cluster routes there. Ask the fabric for a model and it finds a peer that already has it loaded.
How the Network Works
Machines join, declare what they can serve, and the cluster routes on what is actually loaded
Peers, not regions
There is no region menu to choose from. A node discovers its peers over the fabric, and the topology is whatever is connected right now — you can ask it and read the answer.
Metered by the work
A tenth of the block reward funds AI compute, and a provider's share is scaled by the level of the work it did. No subscription, no seat, nothing billed while a machine sits idle.
Proof of AI
A node's confidential-compute tier is attested rather than declared, and what it earns is scaled by that attestation and by a trust score. A claim about a machine is worth nothing on its own.
Whatever the peers have loaded
Ask the cluster which models it serves, which node would answer for one, and where a model nobody has loaded should be placed. Routing follows what is really in memory.
One command to join
hanzo fabric join names a network and starts the node on it. If you would rather lend the machine to builds than to inference, hanzo runner offers the same box as a CI runner.
Anchored, not asserted
Work is ordered on the Hanzo L2 and committed to Lux as a Merkle root, under a stake-weighted validator set. What you can check is what settled, not what a status page says.
Two ends of the same fabric
Take a key and call it, or bring a machine and serve it. Most people who do the second started by doing the first.