ACP0: Agent-Commerce Protocol

The HTTP of AI Shopping / AI 购物的 HTTP 协议

Open standard for AI-to-AI commerce. Let buyer agents and seller agents negotiate directly, no platform needed.

AI 对 AI 购物的开放标准协议。让买家代理与卖家代理直接协商，无需平台中介。

---

⚡ Why ACP0 is Fast (Not Another Slow Blockchain App)

ACP0 is NOT a blockchain e-commerce platform.

It's a high-speed off-chain protocol with optional on-chain anchoring:

• 💨 99% traffic: MQTT/HTTP messaging (< 50ms latency)
• ⛓️ 1% traffic: Blockchain hashes for dispute resolution
• 🚫 0% dependency: Works without any blockchain at all

Think of it as:

• 📡 Broadcast: MQTT (like WhatsApp for agents)
• 🏛️ Trust: Blockchain (like notary, only used when disputes happen)

Example - Buying a $500 laptop:

1. Your agent broadcasts intent via MQTT     →  0ms,  $0
2. 10 sellers respond in 50ms                →  Off-chain
3. You buy with Stripe                       →  Off-chain
4. Optional: Store deal hash on Arbitrum     →  $0.0002

Blockchain is the "stone tablet on the roadside", not the highway.

---

🚀 Current Status: v0.9.0 (MVP)

What's Ready:

• ✅ Complete protocol specification (Intent/Offer/Deal)
• ✅ ECC signature and verification
• ✅ Reference implementation with 32 passing tests
• ✅ Working demo (1-2s transaction time)

What's Coming in v1.0 (Q1 2025):

• 🚧 MQTT network layer (real-time broadcast)
• 🚧 HTTP Indexer (distributed discovery)
• 🚧 Blockchain anchoring (optional trust layer)
• 🚧 Async/await refactoring

What This Means: v0.9.0 proves the protocol works. It's ready for:

• ✅ Protocol discussion and feedback
• ✅ Academic research
• ✅ Building proof-of-concepts
• ❌ Not yet for production e-commerce (wait for v1.0)

See ROADMAP.md for the full plan.

---

🎯 What is ACP0?

The Story Behind It

In November 2024, Amazon sued Perplexity for building "Comet" - an AI agent that logs into Amazon, compares products, and buys based on real user needs instead of paid ads.

Amazon's lawsuit revealed a fundamental conflict:

• Platforms want control: They make money from ads and recommendations
• Users want agents: AI that works for them, not for advertisers

ACP0 is the answer: What if agents don't need platforms at all?

How It Works

┌─────────────┐                                    ┌─────────────┐
│ Buyer Agent │─────── 1. Broadcast Intent ───────▶│  MQTT/HTTP  │
│  (User AI)  │                                    │   Network   │
└─────────────┘                                    └─────────────┘
       ▲                                                  │
       │                                                  │ 2. Forward
       │                                                  ▼
       │                                           ┌─────────────┐
       │                                           │Seller Agent │
       │                                           │Seller Agent │
       │                                           │Seller Agent │
       │                                           └─────────────┘
       │                                                  │
       │                                                  │ 3. Send Offers
       │                                                  ▼
       └───────────── 4. Compare & Select ───────────────┘
                            │
                            │ 5. Payment & Delivery
                            ▼
                      [Transaction Complete]

Core Components:

• Intent: Structured shopping request (JSON)
• Offer: Merchant's response with price, stock, delivery
• Deal: Payment and fulfillment confirmation
• Signature: ECC-based verification (no middleman needed)

---

⚡ Quick Start

Installation

pip install acp0

Buyer Agent (3 lines)

from acp0 import BuyerAgent

agent = BuyerAgent()
offers = agent.broadcast(
    category="laptop",
    budget_range=(4000, 6000),
    currency="CNY",
    location="shanghai"
)

best = agent.select_best(offers)
agent.purchase(best)

Seller Agent (3 lines)

from acp0 import SellerAgent

agent = SellerAgent(
    name="MyShop",
    inventory="./products.json",
    location="shanghai"
)

agent.listen()  # Start receiving intents and auto-respond

That's it! Two agents just completed a transaction without any platform.

Run the Demo

# Clone the repository
git clone https://github.com/acp0/acp0.git
cd acp0

# Install dependencies
pip install -e .

# Run minimal demo
python examples/minimal_demo.py

Expected output:

🚀 ACP0 Minimal Demo Starting...
📦 Setting up Seller Agent...
✓ Seller is listening for Intents...
🛍️ Setting up Buyer Agent...
✓ Buyer is ready
📢 Buyer broadcasting Intent...
📬 Received 1 offer(s)
🏆 Best Offer Selected
💳 Buyer confirming purchase...
✅ Deal confirmed
🎉 Transaction Complete!

---

🌟 Key Features

✅ Platform-Free

• Works with WeChat shops, Shopify stores, 1688 suppliers, or any merchant
• No listing fees, no commission cuts

✅ Agent-First Design

• Structured JSON for AI agents (not keyword search for humans)
• Real-time negotiation between buyer AI and seller AI

✅ Trust Without Middlemen

• Cryptographic signatures verify every message
• Optional blockchain anchoring for disputes
• Decentralized reputation (coming in v1.0)

✅ Lightweight & Extensible

• Core spec: 3 message types, <500 lines of code
• Plug in any payment method (Stripe, Alipay, crypto)
• Add custom fields without breaking compatibility

---

📖 Documentation

Core Documents:

• Protocol Specification - Message formats and security
• Design Philosophy - Why 99% off-chain + 1% on-chain
• Implementation Guide - How to build it
• FAQ - Common questions
• ROADMAP - What's next

For Developers:

• CONTRIBUTING.md - How to contribute
• API Documentation - Python SDK reference
• Examples - Code samples

---

🚀 Use Cases

1️⃣ WeChat Shop Discovery

Problem: 1M+ WeChat shops have great products but no way to be discovered by AI agents
Solution: Shops implement ACP0, agents can find them via intent broadcast

2️⃣ Cross-Border B2B

Problem: Alibaba/1688 suppliers pay high platform fees
Solution: Direct agent-to-agent negotiation, no middleman

3️⃣ Private Domain Commerce

Problem: Influencers sell via manual messages (inefficient)
Solution: Deploy a seller agent, handle inquiries 24/7

4️⃣ AI Shopping Assistants

Problem: Today's AI can only search Amazon/Taobao (controlled by ads)
Solution: Access decentralized merchant network with better prices

---

🛠️ Implementation Status

Component	Status	Notes
Core Protocol	✅ Complete	Intent/Offer/Deal messages
Signatures	✅ Complete	ECC-based verification
Agents	✅ Complete	Buyer and Seller logic
In-Memory Network	✅ Complete	Demo/testing only
MQTT Network	🚧 Planned v1.0	Real-time broadcast
HTTP Indexer	🚧 Planned v1.0	Distributed discovery
Blockchain	🚧 Planned v1.0	Optional anchoring
Smart Contracts	📋 Planned v2.0	Dispute resolution

See ROADMAP.md for detailed plans.

---

🧪 Testing

# Run all tests
pytest tests/ -v

# Run with coverage
pytest tests/ --cov=acp0 --cov-report=term-missing

# Run specific test
pytest tests/test_messages.py -v

Current Test Status:

• ✅ 32 tests passing
• ✅ 100% pass rate
• ✅ Core protocol fully tested

---

🤝 Contributing

We welcome contributions from everyone! Here's how to get started:

1. Read CONTRIBUTING.md - Guidelines and workflow
2. Check Good First Issues - Easy tasks for newcomers
3. Join the discussion - GitHub Issues or Discussions
4. Submit a PR - We review within 48 hours

Priority Areas for v1.0:

• 🔴 MQTT network layer implementation
• 🔴 HTTP Indexer service
• 🔴 Async/await refactoring
• 🟡 Additional language SDKs (Node.js, Rust, Go)
• 🟡 Documentation improvements

---

🌐 Community

Communication Channels

• GitHub Issues - Bug reports, feature requests, RFCs
• GitHub Discussions - General questions and ideas
• Discord - Real-time chat (coming soon)
• Twitter - @acp0protocol
• 中文社区 - 即刻 | 知乎

Monthly RFC Meetings

When: First Saturday of each month, 10:00 AM UTC
Where: Zoom (public, link in Discussions)
What: Protocol discussions and RFC voting

---

🏆 Hall of Fame

🥇 First Production Deployment: [Waiting for you]
🥈 First Cross-Border Transaction: [Waiting for you]
🥉 First 1M GMV Milestone: [Waiting for you]

Want your name here? Ship something awesome with ACP0!

---

📜 About This Project

Created by Human + AI Collaboration

ACP0 was designed by @deloog, a non-technical founder, working with Claude (Anthropic), DeepSeek, and KIMI. This project itself is a proof that:

If humans and AI can design protocols together, imagine what agents can achieve in commerce.

如果人类和 AI 能一起设计协议，想象代理在商业中能成就什么。

Inspiration

This project was inspired by the Perplexity vs Amazon case, which showed that:

• AI agents can challenge traditional platforms
• Users want agents that work for them, not for advertisers
• The future needs open protocols, not closed platforms

---

📄 License

• Protocol Specification: CC0-1.0 (Public Domain) - Use freely, modify freely
• Reference Implementation: MIT License
• Documentation: CC-BY-4.0

---

🔗 Links

• GitHub: https://github.com/acp0/acp0
• Documentation: https://github.com/acp0/acp0/tree/main/docs
• Releases: https://github.com/acp0/acp0/releases
• Issues: https://github.com/acp0/acp0/issues
• Discussions: https://github.com/acp0/acp0/discussions

---

🌟 Star History

---

Built with ❤️ by the ACP0 Community

"The future of shopping is agent-to-agent, not human-to-platform."
"购物的未来是代理对代理，而非人类对平台。"

---

📊 Quick Stats

• Protocol Version: v0.9.0
• Tests: 32 passing
• Transaction Time: 1-2 seconds
• Lines of Code: ~2,000 (core)
• Languages: Python (more coming)
• License: MIT

---

Ready to build the future of commerce? 🚀

Get Started | Read Docs | Contribute | Join Community