SDK

Overview

The 0G Compute Network SDK enables developers to integrate AI inference services from the 0G Compute Network into their applications. Currently, the 0G Compute Network SDK supports Large Language Model (LLM) inference services, with fine-tuning and additional features planned for future releases.

In just five minutes, you can initialize your broker to manage operations, set up and fund your account to pay for services, and learn how to send requests and handle responses.

Features

• Easy integration with AI providers on the 0G Network with built-in billing and account management
• Compatible with OpenAI SDK format
• Support for verifiable responses

TypeScript SDK

Installation

pnpm add @0glabs/0g-serving-broker @types/crypto-js@4.2.2 crypto-js@4.2.0

Quick Start Guide

Initialize the Broker

The broker is your main interface to the 0G Compute Network. It handles authentication, billing, and service communication.

Before initializing the broker, you need:

1. A wallet signer (implements either JsonRpcSigner or Wallet from ethers package)
2. (Optional) The contract address for the 0G Serving contract

Option 1: Using a private key

const { ethers } = require("ethers");
const { createZGServingNetworkBroker } = require("@0glabs/0g-serving-broker");

const provider = new ethers.JsonRpcProvider("https://evmrpc-testnet.0g.ai");

// Get the signer
const privateKey = "INPUT_PRIVATE_KEY_HERE";
const wallet = new ethers.Wallet(privateKey, provider);

// Initialize broker
const broker = await createZGServingNetworkBroker(wallet);

Option 2: Using a browser wallet

import { BrowserProvider } from "ethers";
import { createZGServingNetworkBroker } from "@0glabs/0g-serving-broker";

async function initializeBrokerWithWallet() {
  // Ensure wallet is installed
  if (typeof window.ethereum !== 'undefined') {
    // Create a provider
    const provider = new BrowserProvider(window.ethereum);
    
    // Get the signer
    const signer = await provider.getSigner();
    
    // Initialize broker
    const broker = await createZGServingNetworkBroker(signer);
  }
}

💡 The contractAddress parameter is optional - the SDK uses a default address if not provided.

Discover Available Services

The 0G Compute Network hosts multiple AI service providers. The service discovery process helps you find and select the appropriate services for your needs.

const services = await broker.listService();

Each service contains the following information:

type ServiceStructOutput = {
  provider: string;  // Provider's wallet address (unique identifier)
  name: string;      // Service name
  serviceType: string; // Type of service
  url: string;       // Service URL
  inputPrice: bigint; // Price for input processing
  outputPrice: bigint; // Price for output generation
  updatedAt: bigint;  // Last update timestamp
  model: string;      // Model identifier
  verifiability: string // Indicates how the service's outputs can be verified. 'TeeML' means it runs with verification in a Trusted Execution Environment. An empty value means no verification.
};

Account Management

The 0G Compute Network uses a prepaid account system for each provider. Before using any services, you need to set up and fund an account for each provider you want to use.

Create an Account

await broker.addAccount(providerAddress, initialBalance);

The initialBalance needs to be specified in A0GI units.

Add Funds

await broker.depositFund(providerAddress, amount);

The amount needs to be specified in A0GI units.

Making Service Requests

Service usage in the 0G Network involves two key steps:

1. Retrieving service metadata
2. Generating authenticated request headers

// Get service metadata
const { endpoint, model } = await broker.getServiceMetadata(
  providerAddress,
  serviceName
);

// Generate request headers
const headers = await broker.getRequestHeaders(
  providerAddress,
  serviceName,
  content
);

Note: Headers generated by getRequestHeaders are single-use only & Each request requires new headers

Send Request

You can make requests using either the native fetch API or the OpenAI SDK.

Using fetch:

const response = await fetch(`${endpoint}/chat/completions`, {
  method: "POST",
  headers: {
    "Content-Type": "application/json",
    ...headers,
  },
  body: JSON.stringify({
    messages: [{ role: "system", content }],
    model: model,
  }),
});

Using OpenAI SDK:

const openai = new OpenAI({
  baseURL: endpoint,
  apiKey: "", // Leave empty
});

const completion = await openai.chat.completions.create(
  {
    messages: [{ role: "system", content }],
    model: model,
  },
  {
    headers: {
      ...headers,
    },
  }
);

Response Processing

This function is used to verify the response and settle the fee. If it is a verifiable service, it will return whether the response is valid.

const valid = await broker.processResponse(
  providerAddress,
  serviceName,
  content,
  chatID // Optional: Only for verifiable services
);

Manual Fee Settlement

If processResponse fails, you can settle fees manually by calling settleFee function.

await broker.settleFee(providerAddress, serviceName, fee);

Helper Functions

Get Account Details

This function is used to get account details (like balance) on the given provider address.

const accountDetails = await broker.getAccount(providerAddress);

Request Refund

This function is used to request a refund from your account.

await broker.requestRefund(user, providerAddress, amount);

Other SDKs

Coming soon