ONES1mple $ONE/
Working Draft — v0.2

ONE Protocol

A Token-Gated Decentralized AI Compute Network

Abstract

ONE is a decentralized network for AI inference in which the protocol token is the key to access, not the unit of payment. Holders receive a daily allotment of free compute credits funded by protocol transaction fees rather than by token emissions. Power users buy guaranteed capacity in USDC; GPU operators are paid in USDC. The token captures the value of access and governance while keeping the consumer experience free of per-prompt billing and worker income free of token-price exposure.


1.Introduction

The economics of AI inference are governed by one stubborn fact: every token of output costs real money in GPU time, electricity, and bandwidth. Centralized providers hide this behind flat subscriptions and silent rate limits. Decentralized alternatives usually expose it as raw metered billing, which consumers reject. Neither model is satisfying.

ONE takes a different route by separating two things that are normally fused. Access is granted by holding the token. Settlement happens in USDC. A holder above a stake threshold receives daily credits and never sees an invoice; the compute is paid for from a treasury that the protocol fills out of transaction-fee revenue.

The result is a network with three coupled but independent economic loops: a consumer loop priced in attention and token-holding, a capacity loop priced in USDC, and a supply loop that pays workers in USDC.


2.Network Overview

Three participant classes interact through one job market.

HoldersStake $ONE and consume inference through the Included lane; credits refresh daily and do not roll over.
SponsorsBuy USDC-denominated capacity through the Dedicated lane: priority queueing, every model, reserved throughput.
WorkersSupply GPU capacity, execute jobs, and settle in USDC every six hours.

Every job in either lane emits a signed receipt binding request, model, worker, and result — the shared substrate for verification, billing, and settlement.