Skip to main content

Neuronet State

Everything that defines a Neuronet at a given moment lives on chain: its tasks, its miners, how much is staked on it, and how much of its capacity is in use. This on-chain state is public and updates as the network runs, so anyone can read the current picture of a Neuronet directly from the blockchain.

This page describes what that state contains. For how an Owner sets it up, see Neuronet Configuration and Task Configuration.

What Is Stored​

The blockchain holds compact records rather than raw AI data. For each Neuronet you can read:

  • The Neuronet itself: its identity, owner, status, and settings.
  • Its tasks: each task's model, verification settings, and weight.
  • Its miners: who is registered, their status, proof-of-compute score, reputation, and any penalty flags.
  • Its stake: the total AET staked on the Neuronet, which sets its capacity.
  • Its usage: how much capacity is in use, and the aggregated record of verified work for the current cycle.

What is never stored on chain is the AI work itself. Prompts, responses, model weights, and verification artifacts stay off chain on the miners' machines. Only compact proofs (hashes and Merkle roots) are committed, which keeps the chain small while still letting anyone verify that work happened.

What You Can Read From It​

Because the state is public, the health of a Neuronet can be read directly:

  • Capacity is the staked AET multiplied by the network's capacity rate. See Staking.
  • Utilization is how much of that capacity is being used, which shows whether a Neuronet has spare room or is overflowing.
  • Miner health comes from each miner's status, proof-of-compute score, reputation, and penalty flags.
  • Activity comes from the aggregated batches of verified work submitted each cycle.

How It Differs From a Neurograph​

Older neuronet-based networks expose network state as a single "neurograph" structure built around fixed participant slots (UIDs), validator trust scores, and published IP addresses. AETRON does not work that way:

  • There are no UID slots and no fixed seats, so there is no slot table to read.
  • There is no separate validator role with trust scores, because miners verify each other.
  • Miner endpoints are not published on chain. Routing happens off chain through the Neuronet's request gateway (either the gateway AETRON hosts for you, or the Owner's own self-hosted gateway), so a miner's IP stays private.

Instead of one graph object, the state is a set of on-chain records you read directly. See Chain Reference for the storage items these records live in and the calls that write them. The result is the same kind of transparency, a live and verifiable view of the network, without exposing what does not need to be public.

Last updated on