Hanzo
Proof of AI

Node Operator Guide

Run Hanzo Node, advertise capabilities, receive work, and get paid — plus the trust and TEE attestation requirements, with an honest line between what works today and what is roadmap.

Node Operator Guide

Hanzo Node (github.com/hanzoai/node) lets a machine join the network with a cryptographic identity, advertise what it can do, receive work peer-to-peer, and get paid. This guide describes the model and is explicit about what is live versus roadmap.

Status, plainly. Today a node enrolls, advertises tool and agent offerings, receives requests over a peer-to-peer network, runs them, and settles payment in stablecoins via x402. A fully hardware-metered, TEE- attested GPU marketplace with on-chain DEX settlement is the direction, built on the primitives below — not a shortcut around them.

What a node does today

CapabilityStatusNotes
Enroll with a cryptographic identityLiveEd25519 node identity + registration-code enrollment; resolvable via an on-chain identity registry.
Advertise offeringsLiveYou advertise tools/agents you provide. Hardware (GPU/CPU/RAM) advertisement is roadmap.
Receive work over peer-to-peerLivePaid-work requests arrive as invoice requests over the network.
Run the workLiveThe node executes the requested tool/agent and returns the result.
Get paidLive (x402)Settlement uses the x402 stablecoin protocol via a payment facilitator.
Proof-of-correct-workRoadmapThere is no attestation that the computation was correct yet (see Trust).
Hardware-metered billingRoadmapPricing is per-call today, not per-token / per-FLOP.
On-chain compute DEX settlementRoadmapOrder-book/DEX settlement is not in the shipped build.

Running a node

The node is a Rust binary. Because exact flags evolve, follow the canonical README in hanzoai/node; the model is:

  1. Build/install the node binary from the repository.
  2. Initialize identity. On first run the node generates its keypair and completes registration-code enrollment, establishing the node's identity.
  3. Configure a wallet. Settlement is in stablecoins over x402, so the node needs a receiving wallet/payment address.
  4. Publish offerings. Declare the tools/agents you provide and their price (per-use, or free). These become discoverable to requesters.
  5. Stay online. The node listens for invoice requests, runs accepted work, returns results, and settles payment.

Getting paid

  • Rail: stablecoin settlement via x402. A requester pays the invoice; the facilitator settles to your receiving address after the work runs.
  • Pricing: per-invocation today (a flat price per tool call, or free). Token- and FLOP-based metering is roadmap.
  • What is not wired yet: raw on-chain transfer paths and the queue-driven auto-fulfillment loop are still being completed; treat large-scale unattended settlement as roadmap and reconcile payouts you receive.

Trust and TEE attestation

Two different trust models apply depending on where the work runs.

On-chain verifiable inference — no special hardware required. Small models that run inside the inference precompile (0x0300...0003) are trustless by recomputation: every validator re-runs the identical integer computation. You do not need a TEE for this path. See Proof of AI.

Off-chain compute on operator hardware — attestation is the anchor. For models too large to recompute on-chain, the trust anchor is a hardware attestation proving your machine ran a genuine, unmodified workload.

AttestationStatus
AMD SEV-SNPSupported — implemented end-to-end (full certificate-chain verification against AMD's key service). This is the day-one attestation target for confidential operators.
Intel TDXRoadmap (scaffolded, not finished).
NVIDIA GPU confidential compute (H100/H200/Blackwell)Roadmap (scaffolded, not finished).

On-chain attestation precompiles exist at 0x0301 (GPU), 0x0302 (TPM), and 0x0303 (compute) as the verification sink for these as they harden.

Operator posture today. Because attested GPU compute is not finished, the production posture for sensitive workloads is vetted operators plus SEV-SNP where available, with the on-chain recomputation path used wherever a model is small enough to verify directly. Do not assume your off-chain results are independently attested yet.

Roadmap summary

  • Hardware capability advertisement and per-token/FLOP metering.
  • Proof-of-correct-work for off-chain compute (attestation-bound).
  • Finished TDX and NVIDIA GPU confidential-compute attestation.
  • On-chain marketplace/DEX settlement for compute orders.

How is this guide?

Last updated on

On this page