0friction

Cross-chain AI compute via 0G. Pay with USDC, no bridge, no 0G token management for end users.

0friction is a framework-level project for the 0G Open Agents track:

• @0friction/sdk (TypeScript SDK for any runtime)
• Solver backend (0G execution + USDC settlement)
• Demo frontend and working CLI agent example

Live Links

• Live backend (Render): https://zerofriction-solver.onrender.com
• Live frontend (Vercel): https://0friction.vercel.app
• npm package: https://www.npmjs.com/package/@0friction/sdk
• GitHub: https://github.com/officialcmg/0friction

Architecture Diagram

Why this matters

0G compute is powerful, but user onboarding breaks when apps require bridging and native token flow. 0friction abstracts that away:

• users sign gasless auth on home chain
• solver executes inference on 0G
• solver settles USDC on home chain

This turns 0G compute into a drop-in developer primitive for agents and apps.

Project Structure

0friction/
├── sdk/              @0friction/sdk (published npm package)
├── backend/          Solver backend (Express + 0G compute SDK)
├── frontend/         Next.js demo app
├── examples/         Working example agent (CLI)
├── contracts/        IntentRegistry (deployed on 0G Galileo)
└── architecture-0friction.svg

Deployment and Contracts

• IntentRegistry (0G Galileo / chain 16602):
  • 0x01D1084d915eAb33A36FBaBFC29Dc8e6478b0926
  • https://chainscan-newton.0g.ai/address/0x01D1084d915eAb33A36FBaBFC29Dc8e6478b0926
• Solver address:
  • 0xB9a33C169d1360E6AdFf7266797f85467856bCc2
• Home-chain payment token (Base Sepolia USDC):
  • 0x036CbD53842c5426634e7929541eC2318f3dCF7e

SDK usage (high level)

import { createClient } from "@0friction/sdk";

const client = createClient();

const quote = await client.quote.get({
  model: "qwen/qwen-2.5-7b-instruct",
  prompt: "Explain 0G compute in one sentence.",
});

const intent = client.intent.build({
  quote,
  owner: userAddress,
  requestPayload: {
    model: "qwen/qwen-2.5-7b-instruct",
    messages: [{ role: "user", content: "Explain 0G compute in one sentence." }],
  },
  nonce: String(Date.now()),
});

// App layer signs intent + permit using viem/ethers/wagmi
const result = await client.intent.submit({
  intent,
  intentSignature,
  permit,
  permitSignature,
  requestPayload,
  quoteId: quote.quoteId,
});

Gasless flow (v1)

For each request:

1. SDK fetches quote from solver.
2. User signs EIP-712 ComputeIntent (gasless signature).
3. User signs EIP-2612 USDC Permit (gasless signature).
4. Solver verifies, runs compute on 0G, and settles via permit + transferFrom.
5. App receives response + audit bundle.

Working example agent

• CLI agent path: examples/cli-agent.ts
• SDK E2E test path: test-sdk.ts

0G features used

• 0G Compute inference via @0gfoundation/0g-compute-ts-sdk
• Service discovery (listService)
• TEE response verification (processResponse)
• 0G chain deployment (IntentRegistry contract)

Current scope and roadmap

This is an initial version optimized for hackathon reliability:

• v1: Base Sepolia + USDC + per-request signatures
• Next: multi-chain settlement and smoother session-style auth (no per-message signing UX)
• Next: richer agent modules and deeper OpenClaw integrations

Quick local run

npm install
cp .env.example .env
npm run dev --workspace=backend
# in another terminal
cd frontend && npm run dev

Team

• Chris Gachau
• X: chrismgtweets_
• Telegram: chrismgeth

License

MIT