Skip to the playground

HTTP

GPUs that agents
pay for themselves.

One HTTP request rents a Bittensor Subnet 51 GPU from Lium. The agent pays with x402, in USDC or TAO. No account, no API key, no human.

Rent one from here
$npm i @gpu402/tools

— Lium nodes, live. Drag the globe.

Live
—
Paid
—
GPU min
—

01 —— How it works / Step 1 · Quote

Ask for a GPU. Get a price.

The agent calls the gateway. It answers HTTP 402 with the price of a 15-minute block, taken from Lium’s live market.

gpu402 / agent

  1. $ pnpm demo:agent
  2. Waiting for the first agent to pay▍

Every GPU here is a real machine, on a Subnet 51 miner.

nodes
—
GPU models
—
min blocks
15

Drag your pointer across it. The fans spin up while a pod is live.

02 —— Playground

Don’t take our word. Send it.

Every button here hits the live gateway. Price a GPU, read its 402, then rent one with your own wallet.

Time
15 min

Ask the gateway what this GPU costs right now. Nothing is paid.

The real response appears here.

import { Gpu402Gateway } from "@gpu402/tools";

const gw = new Gpu402Gateway({
  url: "https://calibration-dee-various-cooking.trycloudflare.com",
  wallet: process.env.AGENT_PRIVATE_KEY, // USDC, no gas
  payWith: "usdc", // or "tao"
});

const quote = await gw.quote({ gpu: "RTX 4090", blocks: 1 });
const { data: pod, payment } = await gw.rent({
  gpu: "RTX 4090", blocks: 1,
  sshPublicKey, // your OpenSSH public key
});
const ready = await gw.waitReady(pod.podId);
console.log(ready.ssh.command, payment.txHash);

Updates as you pick a GPU and time on the left.

04 —— Packages

Two packages. One line to a GPU.

Live from the npm registry. Use the SDK with your Lium key, or give the tools to an agent and let it pay.

The SDK

gpu402

npm

Rent the cheapest Lium GPU, run a job over SSH, stop it. Inside a budget.

  • gpu402Lium client: listGpus, quote, rentCheapest, waitReady, exec, stop
  • gpu402/openapiEvery Lium endpoint, typed from its OpenAPI spec
$npm i gpu402

Tools for agents

@gpu402/tools

npm

Four tools and a wallet. Your agent rents, runs and pays per use.

  • @gpu402/tools/aiVercel AI SDK tools
  • @gpu402/tools/langchainLangChain.js tools
  • @gpu402/toolsGpu402Gateway, the x402 client
  • @gpu402/tools/taoPay in TAO on Bittensor EVM
$npm i @gpu402/tools

Self-host

The gateway

The x402 server behind this page. Run your own with a Lium key and a payout address.

  • Takes USDC or TAO
  • Refunds failed rentals
  • Removes pods when paid time ends
$pnpm gateway

05 —— API

Plain HTTP, all the way down.

Pick an endpoint and send it. The free ones answer right here.

06 —— Network

Compute from Subnet 51.

Bittensor pays Lium’s miners to serve GPUs. GPU402 puts a price and a wallet in front.

Subnet 51 · Bittensor Finney

Market · GPU402 price per 15 minutes

07 —— Live / Activity

What the gateway is doing now.