Architecture

How YoxoAI is built

The protocol is split into four independent layers so each can evolve without breaking the others. Below is the current testnet design — details will change as the network hardens toward mainnet.

Layer 1

Node coordination

A libp2p-based gossip network lets GPU nodes advertise capacity, price, and hardware specs. Jobs are matched off-chain and settled on-chain.

Layer 2

Proof of compute

Nodes submit succinct proofs of completed work using a mix of redundant sampling and zk-SNARK circuits for high-value jobs, so clients don't have to trust node operators blindly.

Layer 3

Settlement & escrow

Payments are escrowed in $YOXO or USDC when a job starts and released automatically once proofs are verified on-chain, with disputes handled by a bonded arbiter set.

Layer 4

Model registry

An on-chain registry tracks model lineage — base weights, fine-tune diffs, licensing — so provenance is verifiable and attribution/royalties can flow back to original authors.

Network specs

Consensus: Proof-of-Stake (validator set of 100)
Settlement chain: Ethereum L2 (OP Stack)
Compute proof: Redundant execution + zk-SNARK for GPU jobs
Node client: Rust, open-source, MIT licensed
Supported workloads: Training, fine-tuning, batch inference
Min node spec: 16GB VRAM, 100 Mbps uplink