Hanzo Node

Consumer cards, datacenter accelerators, or Apple Silicon. The node reports what it can serve and the cluster routes to it on that basis.

A tenth of the block reward is set aside for AI compute. A provider's share is scaled by the level of the work — embeddings and small models, chat-sized models, large and multimodal, training, or specialized compute like proofs.

Ask it for models and it lists everything served across the peers your node can see. Ask it to route and it names who would answer; ask for placement and it says where a model nobody serves should be loaded.

A node's confidential-compute tier is attested rather than declared, and its share is scaled by that tier and by a trust score. Claiming a capability is not the same as proving one.

up starts it on the active network, join switches network and starts it, stop signals the process this CLI started by its recorded pid — and never anything else that happens to share the name.

Work is ordered on the Hanzo L2 and committed to Lux as a Merkle root, so what a node did is anchored where it cannot quietly be rewritten later.