Skip to main content

Core Operating Principles

These are the building blocks of how AETRON works. Understanding them makes the rest of the documentation easier to follow.

Neuronets​

A Neuronet is an AI service run by distributed infrastructure: a self-contained marketplace for one or more AI tasks where miners produce results and the network verifies them. The network can run many independent Neuronets at once across different task classes.

How a Neuronet Is Structured​

Each Neuronet brings together:

  • An Owner who creates it, defines its tasks and parameters, and runs the service.
  • Miners who provide compute, serve requests, and re-check each other's work.
  • Stakers who delegate AET to it, which raises its capacity and earns them a share of its emission.

A single Neuronet can host several tasks at once (for example a language model, an image model, and a speech model) and attract any number of miners.

What Neuronets Can Do​

Neuronets can specialize in a wide range of AI work:

  • Generating and processing text
  • Image and video generation
  • Speech recognition and audio processing
  • Document analysis and data processing
  • Inference and model training
  • AI agents that use tools and automate workflows

How Trust Works: Proof of Intelligence​

The central idea of AETRON is that every computation is verified cryptographically rather than scored by opinion.

  • Miners produce a proof that a computation ran exactly as declared: same model, same parameters, same hardware.
  • Other miners selectively re-execute a portion of the work and compare results. Challenges are indistinguishable from ordinary user traffic, so a miner cannot behave well only when watched. See Shadow Replay.
  • Emission is paid only for work that passes verification. A failed check means no reward, plus a loss of reputation and future task share.

Training runs are proven differently from single requests, since there is no one output to compare; see Proof of Training.

There is no separate "validator" role that subjectively rates miners. Verification is a job that miners do for each other, enforced by the protocol. See Proof of Intelligence for the details.

Roles in the Ecosystem​

AETRON has four roles, and one participant can hold several at once.

Owner​

Creates a Neuronet, sets its tasks and rules, and operates the service. The Owner receives a fixed share of the Neuronet's emission (a creator fee that can be lowered but not raised) and decides how to monetize the service for end users.

Miner​

Runs the hardware, executes inference or training, and serves requests. Miners also verify each other: the protocol assigns a share of work to be re-checked. They earn emission for verified work, weighted by their performance and reputation.

Staker​

Delegates AET to a specific Neuronet. A stake does two things at once: it votes for the usefulness of that service and it increases the service's capacity. Stakers earn a fixed share of the Neuronet's emission in proportion to the amount they stake.

Block Validator​

Secures the base blockchain, produces blocks, and votes on protocol upgrades. This role does not perform AI verification and does not need a GPU. It earns a separate share of emission for maintaining the network.

Tokens and Economics​

The AET Token​

AET is the native token of the network. It is used for staking, registration fees, and transaction fees. There is no protocol-level payment for inference: AET reaches participants through emission for verified work.

  • Hard cap of 33 million AET, with no premine.
  • Emission of 1.5 AET per block at a 12-second block time, about 10,800 AET per day in the current epoch.
  • Halving by total issuance, with the first halving at 16.5 million AET (half of the cap). The exact date shifts because recycled fees return tokens to the unissued pool.

How Emission Is Split​

Each block's emission is divided at the protocol level:

  • 90% goes to Neuronets, shared by how much verified useful work each one produced.
  • 5% goes to Block Validators for securing the network.
  • 5% goes to the Treasury for development, audits, and grants.

Inside a Neuronet, the share is split again between its miners, its owner, and its stakers.

Staking and Capacity​

Staking AET into a Neuronet sets how much work it can be paid for. More stake means more capacity, which means more requests can earn emission. Stake is locked, not burned, and returns after a cooldown period. Stake without real usage earns little, so capacity only has value when paired with genuine demand.

Governance​

AETRON moves from team-run authority toward decentralized control in stages.

  • The Neuronet layer is open from the start: no central party can block the creation of a Neuronet, remove a miner, or censor requests.
  • The base blockchain transitions to on-chain voting where only Block Validators vote, with one validator counting as one vote. This keeps large token holders from dominating decisions by weight of stake alone.
  • Protocol upgrades are proposed, voted on, and enacted on chain once they reach quorum.

Network Architecture​

AETRON separates settlement from execution.

On-chain (settlement)​

The blockchain stores hashes, Merkle roots, scores, and balances. It records what happened and enforces the rules, but it does not run AI workloads itself.

Off-chain (execution)​

AI requests, responses, model weights, and verification artifacts live off chain on the miners' machines. Only compact proofs are committed on chain, which keeps the blockchain small while supporting heavy AI workloads.

These principles work together to make AETRON a network where rewards follow proven, useful work rather than reputation or opinion.

Last updated on