Skip to main content

Neuronet Configuration

A Neuronet is configured by its Owner. Some settings apply to the Neuronet as a whole, while others belong to the individual tasks it runs. This page covers the Neuronet-level settings. For the per-task settings (model, verification, privacy, and economics of a single job), see Task Configuration.

Who Can Change What​

  • The Owner controls the Neuronet's own settings and the configuration of its tasks.
  • Some parameters are network-level, governed by the protocol rather than set by an individual Owner. The registration fee schedule and the minimum activation stake are examples: they are the same across the network and change only through governance.
  • Everything is publicly readable on chain, so anyone can inspect how a Neuronet is set up.

Neuronet-Level Settings​

These apply to the whole Neuronet, across all of its tasks.

  • Identity. A name and descriptive metadata for the Neuronet, which the Owner can set and clear.
  • Tasks. The Owner adds and removes tasks. Each task has its own configuration; see Task Configuration.
  • Task weights. When a Neuronet runs several tasks, weights decide how its emission is divided among them.
  • Owner fee. The Owner's share of emission. It can be lowered but never raised.
  • Discovery and gateway. Routing details that tell the network how requests reach the Neuronet's miners, including the gateway address for self-hosted setups. See Gateway API for the interface that address has to serve.
  • Miner access. For Confidential (TEE) Neuronets the Owner approves, invites, or removes miners directly. In standard public Neuronets miners join on their own once they qualify.

Capacity and Activation​

A Neuronet's throughput is not a fixed parameter. It comes from stake:

  • Capacity is set by how much AET is staked on the Neuronet, so the Owner raises it by attracting stake rather than by turning a dial.
  • Activation. A Neuronet stays dormant until it reaches the minimum activation stake, after which it begins earning capacity. This minimum is a network-level setting.

See Staking for how stake becomes capacity, and How to Stake for the calls that put it there.

What Is Not Configurable​

Unlike older neuronet-based networks, a Neuronet has no slot counts to tune. There is no cap on the number of miners and no fixed split of validator and miner seats, because AETRON has no slots. There is also no separate validator role to configure: miners verify each other, and the protocol sets when and how. See Proof of Intelligence for how verification is organized.

Changing Settings Safely​

  • Test changes before applying them to a live, well-used Neuronet.
  • Remember that some settings only move in one direction, such as the Owner fee, which can only be lowered.
  • Fixed-for-life settings live at the task level (the model and its canonical execution settings) so that verification stays consistent. See Task Configuration.
Last updated on