Skip to main content

0G Compute SDK

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 inference services, and learn how to send inference requests and handle responses.

Quick Start​

Installation​

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

Core Concepts​

1. The Broker​

Your interface to the 0G Compute Network:

  • Handles authentication and billing
  • Manages provider connections
  • Verifies computations

2. Providers​

GPU owners offering AI services:

  • Each has a unique address
  • Set their own prices
  • Run specific models

3. Prepaid Accounts​

  • Fund account before usage
  • Automatic micropayments
  • No surprise bills

Step-by-Step Guide​

Initialize the Broker​

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

const provider = new ethers.JsonRpcProvider("https://evmrpc-testnet.0g.ai");
const wallet = new ethers.Wallet(process.env.PRIVATE_KEY!, provider);
const broker = await createZGComputeNetworkBroker(wallet);

Fund Your Account​

// Add 0.1 OG tokens (~10,000 requests)
await broker.ledger.addLedger(ethers.parseEther("0.1"));

// Check balance
const account = await broker.ledger.getLedger();
console.log(`Balance: ${ethers.formatEther(account.balance)} OG`);

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.

🎯 Official 0G Services
ModelProvider AddressDescriptionVerification
llama-3.3-70b-instruct0xf07240Efa67755B5311bc75784a061eDB47165DdState-of-the-art 70B parameter model for general AI tasksTEE (TeeML)
deepseek-r1-70b0x3feE5a4dd5FDb8a32dDA97Bed899830605dBD9D3Advanced reasoning model optimized for complex problem solvingTEE (TeeML)
const services = await broker.inference.listService();

Each service contains the following information:

type ServiceStructOutput = {
  provider: string; // Provider's wallet address (unique identifier)
  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.
};

Acknowledge Provider​

Before using a service provided by a provider, you must first acknowledge the provider on-chain by following API:

await broker.inference.acknowledgeProviderSigner(providerAddress)

The providerAddress can be obtained from from service metadata. For details on how to retrieve it, see Discover Available Services

Service Requests​

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

  • Retrieving service metadata
  • Generating authenticated request headers
  
  // Get service details
  const { endpoint, model } = await broker.inference.getServiceMetadata(provider);
  
  // Generate auth headers (single use)
  const headers = await broker.inference.getRequestHeaders(provider, question);

Send a Request to the Service​

const response = await fetch(`${endpoint}/chat/completions`, {
    method: "POST",
    headers: { "Content-Type": "application/json", ...headers },
    body: JSON.stringify({
      messages: [{ role: "user", content: question }],
      model: model,
    }),
  });
  
const data = await response.json();
const answer = data.choices[0].message.content;

Response Processing​

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

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

Fee Settlement​

Fee settlement by the broker service occurs at scheduled intervals.

Account Management​

Check Balance​

const ledger = await broker.ledger.getLedger();
console.log(`
  Balance: ${ethers.formatEther(ledger.balance)} OG
  Locked: ${ethers.formatEther(ledger.locked)} OG
  Available: ${ethers.formatEther(ledger.balance - ledger.locked)} OG
`);

Add Funds​

// Add more funds
await broker.ledger.depositFund(ethers.parseEther("0.5"));

Request Refund​

// Withdraw unused funds
const amount = ethers.parseEther("0.1");
await broker.ledger.retrieveFund("inference", amount);

Troubleshooting​

Common Issues​

Error: Insufficient balance

Your account doesn't have enough funds. Add more:

await broker.ledger.addLedger(ethers.parseEther("0.1"));
Error: Headers already used

Request headers are single-use. Generate new ones for each request:

// ❌ Wrong
const headers = await broker.inference.getRequestHeaders(provider, content);
await makeRequest(headers);
await makeRequest(headers); // Will fail!

// ✅ Correct
const headers1 = await broker.inference.getRequestHeaders(provider, content);
await makeRequest(headers1);
const headers2 = await broker.inference.getRequestHeaders(provider, content);
await makeRequest(headers2);
Error: Provider not responding

The provider might be offline. Try another:

// Try all official providers
for (const [model, provider] of Object.entries(OFFICIAL_PROVIDERS)) {
  try {
    console.log(`Trying ${model}...`);
    return await makeRequestToProvider(provider);
  } catch (e) {
    console.log(`${model} failed, trying next...`);
    continue; // Try next provider
  }
}

Next Steps​

Questions? Join our Discord for support.