Introduction

Nodus is an AI compute coordination layer powered by the ICPX Protocol. It gives developers, agents and applications one simple, scalable interface to decentralized GPU infrastructure — abstracting away availability, cost management and deployment complexity.

You request compute by type — say an H100 — and Nodus handles discovery, scheduling, billing and teardown, settling usage on-chain through ICPX.

Quickstart

Install the SDK, set your key, and schedule your first job on a decentralized H100.

# 1 · install
npm install @nodus/sdk

# 2 · authenticate
export NODUS_KEY=ndk_live_…

# 3 · run a job
nodus compute run --gpu H100 --image vllm/vllm:latest

Authentication

All requests authenticate with an API key created in the dashboard under API Keys. Keys carry your wallet's spend authority — keep them secret and never commit them to source.

import { Nodus } from "@nodus/sdk";

const nodus = new Nodus({
  apiKey: process.env.NODUS_KEY,
});

Compute Jobs

A job is a containerized workload scheduled onto a GPU. Specify the GPU type, an image and a command — Nodus provisions, runs and tears down automatically, returning a live endpoint.

const job = await nodus.compute.run({
  gpu:     "H100",
  image:   "vllm/vllm:latest",
  command: "python serve.py --model llama-3-70b",
});

console.log(job.endpoint); // https://…nodus.run

Agent Runtime

Agents are long-lived workloads with persistent memory and their own GPU allocation. Deploy once and the runtime keeps them alive, schedulable and observable from the dashboard.

const agent = await nodus.agents.deploy({
  name:    "market-sentinel",
  gpu:     "A100",
  runtime: "node20",
  entry:   "./agent.js",
});

Models

Test and benchmark models in the Playground, then promote them to a managed inference endpoint. Common open models are available preloaded.

API Reference

The REST and SDK surfaces mirror each other. Core resources:

Architecture

Requests flow from your application down to physical GPUs, coordinated by Nodus and settled by ICPX.

Pricing

Pay per second of compute in SOL. No minimums, no idle charges — billing stops the moment a job ends.