A permissionless mesh of 127,000+ edge nodes that serves and verifies AI inference at sub-millisecond latency. Every output attested by zero-knowledge proofs. Every fee settled on Ethereum.
ax0n's architecture decomposes inference into three independent, auditable layers — routing, attestation, and settlement.
Latency-aware dynamic routing engine. Scores every node in the mesh by hardware tier, current stake weight, and historical p95 latency. Selects an optimal cluster in under 2ms with sub-linear lookup via our proprietary Bloom-trie index.
Custom Groth16 circuit verifies model integrity and output determinism without ever exposing weights. Proof size is a constant 288 bytes regardless of model parameter count. Generation overhead: ~120ms on RTX 4090-class hardware.
Byzantine fault-tolerant consensus across 10,000+ active validators using optimistic aggregation with ZK fallback for disputed proofs. Achieves finality in 1.2s under normal conditions, 6s worst-case.
Every inference on ax0n follows a four-stage lifecycle — from client broadcast to on-chain finality.
Client broadcasts a signed inference request to the ax0n mempool, specifying model ID, max acceptable latency, and ax0n fee cap. NeuralRoute selects the optimal node cluster within 2ms using the Bloom-trie index.
The selected node runs inference locally via the ax0n runtime, then passes the output through ProofCore. A 288-byte Groth16 SNARK is generated attesting to both model integrity and output determinism.
A randomly sampled quorum of 256 validators verifies the proof using local verifier keys. The aggregate quorum signature and proof are bundled into the next settlement batch via MeshSync.
Batches settle on Ethereum every 512 blocks. ax0n fees distributed in the same transaction: 78% to node operators, 12% to validators, 10% to protocol treasury.
Core protocol parameters set by governance. All changes subject to a 7-day timelock enforced by the ax0n Governor contract.
ax0n is the staking and fee token of the ax0n protocol. Stake to run a node, earn fees, and vote on governance proposals.
Any machine with a consumer GPU and 1,000 ax0n staked can join the mesh today. Operators earn 78% of all inference fees routed through their cluster.