0G Orbit
The developer toolkit for 0G. Store, infer, fine-tune — one SDK, one CLI, five minutes.
What is this?
0G Orbit wraps the 0G storage SDK, serving broker, and chain interactions into a single unified interface. Instead of juggling @0gfoundation/0g-ts-sdk, @0glabs/0g-serving-broker, and raw ethers calls, you get one Orbit class and one orbit CLI.
Storage — Upload and download files to 0G's decentralized storage network Inference — Run AI models on 0G's decentralized compute network Fine-Tuning — Upload datasets, create training tasks, download fine-tuned models
Install
npm install 0g-orbit
Quick Start
SDK
import { Orbit } from '0g-orbit'
const orbit = await Orbit.connect({
network: 'testnet',
privateKey: process.env.PRIVATE_KEY,
})
// Upload a file
const { root, txHash } = await orbit.store('./data.json')
console.log(`Stored: ${root}`)
// Download it back
await orbit.retrieve(root, './downloaded.json')
// Run AI inference
const response = await orbit.infer('meta-llama/Llama-3.2-3B-Instruct', {
message: 'Explain zero-knowledge proofs in one sentence.',
})
console.log(response.content)
// Fine-tune a model
const dataset = await orbit.uploadDataset('./training-data.jsonl')
const task = await orbit.createFineTuneTask({
model: 'base-model',
dataset: dataset.root,
providerAddress: '0x...',
})
CLI
# Set your private key
export PRIVATE_KEY=0x...
# Upload a file
orbit store ./my-file.txt
# Download a file
orbit retrieve <rootHash> ./output.txt
# Run inference
orbit infer meta-llama/Llama-3.2-3B-Instruct -m "Hello, 0G!"
# List available AI services
orbit services
# Fine-tune a model
orbit fine-tune ./dataset.jsonl --model base-model --provider 0x...
# Check account status
orbit status
API
Orbit.connect(config)
Creates a connected Orbit instance.
| Option | Type | Description |
|---|---|---|
network | 'testnet' | 'mainnet' | Network to connect to |
privateKey | string? | Wallet private key (falls back to PRIVATE_KEY env var) |
rpcUrl | string? | Custom RPC URL |
Storage
| Method | Description |
|---|---|
orbit.store(path, options?) | Upload a file, returns { root, txHash } |
orbit.storeData(text, options?) | Upload a text string |
orbit.retrieve(rootHash, outputPath, options?) | Download a file by root hash |
Inference
| Method | Description |
|---|---|
orbit.infer(model, options) | Run AI inference, returns { content, tokensUsed } |
orbit.listServices() | List available AI services on the network |
Fine-Tuning
| Method | Description |
|---|---|
orbit.uploadDataset(path) | Upload a training dataset to storage |
orbit.createFineTuneTask(options) | Create a fine-tuning task |
orbit.getFineTuneTask(provider, taskId) | Check task status |
orbit.downloadModel(provider, taskId, outputDir) | Download a fine-tuned model |
orbit.listModels() | List available base models |
orbit.listProviders() | List fine-tuning providers |
Direct Client Access
For advanced usage, access the underlying clients directly:
orbit.storage // StorageClient
orbit.inference // InferenceClient
orbit.fineTuning // FineTuningClient
Security
This package has transitive dependencies with known vulnerabilities (from upstream 0G SDKs). To mitigate, add this to your project's package.json:
{
"overrides": {
"axios": "1.14.0"
}
}
See SECURITY.md for details.
Contributing
See CONTRIBUTING.md.