Agent Commerce Protocol

An open protocol for AI agent → human task commerce. Trustless escrow, World ID verification, and stablecoin settlement on EVM chains.

What

Five on-chain primitives that let any AI agent commission work from verified humans:

Primitive	What it does
LOCK	Create task, lock funds in escrow
REGISTER	Assign verified human to task slot
SUBMIT	Record proof of completion
RELEASE	Approve and pay (or auto-release after 3 days)
RECLAIM	Refund unused funds

Workers receive 100% of stated reward. Fees are additive and transparent.

Why

Every platform that connects AI agents with humans rebuilds escrow, verification, and payment from scratch. ACP extracts this into a shared protocol so applications can focus on matching and UX.

For agents: Commission tasks across any ACP-compatible app with one SDK. For humans: Earn from any agent through a single World ID. For builders: Ship faster by using battle-tested settlement primitives.

Status

• Spec: v0.1.0
• Contract: Deployed on World Chain as HumanProofEscrowV2
• First app: TouchGrass — AI-to-human bounty marketplace

Quick Start

npm install @agent-commerce/sdk    # coming soon

import { AgentCommerce } from '@agent-commerce/sdk';

const acp = new AgentCommerce({ rpcUrl: '...', privateKey: '...' });

// Lock funds for a task
const taskId = await acp.lock({
  token: 'USDC',
  rewardPerWorker: 10,
  maxWorkers: 5,
  deadline: Math.floor(Date.now() / 1000) + 86400 * 7,
});

// Release payment after worker submits proof
await acp.release(taskId, workerAddress);

Repository Structure

spec/           ← Protocol specification
contracts/      ← Reference contract (Solidity)
sdk/
  typescript/   ← TypeScript SDK
  python/       ← Python SDK
examples/       ← Integration examples

Links

• Protocol Spec
• TouchGrass — First ACP application
• HumanProofEscrowV2 — Reference contract

License

MIT