logoGPU/docs
Dashboard← Back to site

Introduction

GPU is an open, OpenAI-compatible inference network. Anyone can serve compute and earn Season 0 points; anyone can consume the API with the same SDK they already use.

What is GPU Network

GPU Network is a decentralized GPU inference mesh. Two sides participate. Providers run a small node on any GPU-equipped machine (Mac, Linux box, or even a browser tab) and earn GPU Points for every completed task. Consumers send OpenAI-compatible chat-completion requests and get streamed responses through the mesh.

Why it matters

Most AI inference today is concentrated in a handful of hyperscalers. Outages, rate limits, and policy changes ripple to every dependent product. Meanwhile, hundreds of millions of consumer GPUs sit idle outside business hours. GPU routes the unused supply directly to the demand and scores every contribution as Season 0 points, bound to the provider's wallet.

How it works

  • Consumer → POST /v1/chat/completions with a bearer key
  • Orchestrator → queues the task and selects a live provider that has the requested model
  • Provider → runs inference locally (Ollama / WebGPU) and streams chunks back
  • Provider accrues GPU Points; consumer pays in credits
  • Points convert to API credits any time; lifetime score ranks the leaderboard
Honest disclosure
There is no token. Rewards are Season 0 points — an off-chain contribution score tied to your wallet. Points convert to API credits today and may qualify for future reward programs; they have no guaranteed monetary value and cannot be bought.

Project status

  • OpenAI-compatible API with SSE streaming — shipped
  • Wallet-native auth via MetaMask / WalletConnect — shipped
  • Provider via curl /install | sh — shipped
  • Browser nodes via WebGPU — shipped
  • Multi-model catalogue + nodes-available routing — shipped
  • Daily faucet + points→credits conversion — shipped
  • Season 0 leaderboard — shipped
  • Split-model inference + on-chain settlement (optional) — roadmap
Next →
Install