langchain-amorce 0.1.0

Secure LangChain agents with Amorce (Ed25519 + HITL + A2A)

langchain-amorce

Secure LangChain agents with Amorce in 2 lines of code

Add Ed25519 signatures, human-in-the-loop approvals, and A2A compatibility to any LangChain agent.

---

🚀 Quick Start

Installation

pip install langchain-amorce

Basic Usage (2 Lines!)

from langchain_amorce import AmorceAgent
from langchain.chat_models import ChatOpenAI
from langchain.tools import DuckDuckGoSearchRun

# Create secure agent
agent = AmorceAgent(
    llm=ChatOpenAI(model="gpt-4"),
    tools=[DuckDuckGoSearchRun()],
    secure=True  # ← Adds Amorce security
)

# Use normally - all actions are now cryptographically signed
result = agent.run("What's the latest AI news?")
print(f"Agent ID: {agent.agent_id}")  # Verified identity

That's it! Your LangChain agent now has:

• ✅ Ed25519 cryptographic signatures
• ✅ Trust Directory verification
• ✅ Automatic identity management
• ✅ A2A-compatible messages

---

🛡️ Features

Cryptographic Signatures

Every agent action is signed with Ed25519, providing non-repudiable proof of authorization.

Human-in-the-Loop (HITL)

Require human approval for sensitive operations:

from langchain.tools import ShellTool

agent = AmorceAgent(
    llm=ChatOpenAI(),
    tools=[
        DuckDuckGoSearchRun(),  # No approval needed
        ShellTool()              # Dangerous!
    ],
    hitl_required=['shell'],  # Require approval for shell
    secure=True
)

# Search works normally
agent.run("Search for Python tutorials")

# Shell command triggers approval popup
agent.run("Delete tmp files")  # ← Human approval required!

A2A Protocol Compatible

Messages are formatted for Agent-to-Agent Protocol compatibility:

agent = AmorceAgent(
    llm=ChatOpenAI(),
    tools=[search],
    a2a_compatible=True  # ← Enable A2A format
)

# All messages use A2A envelope with Amorce signatures

---

📖 Examples

Marketplace Demo (Sarah - Buyer Agent)

See the full Enhanced Marketplace Demo showing Sarah negotiating with Henri.

from langchain_amorce import AmorceAgent
from langchain.chat_models import ChatOpenAI

sarah = AmorceAgent(
    name="Sarah",
    role="Smart Shopper",
    llm=ChatOpenAI(model="gpt-4"),
    tools=[
        brave_search_tool,      # Price research
        budget_calculator,       # Affordability check
        fraud_detector,         # Verify seller
    ],
    hitl_required=['make_payment', 'share_address'],
    max_budget=500,
    secure=True
)

# Sarah autonomously:
# 1. Searches for MacBook Pro
# 2. Finds sellers in Trust Directory
# 3. Verifies reputations
# 4. Negotiates price
# 5. Requests human approval for payment

---

🔧 Advanced Configuration

Custom Identity

from amorce import IdentityManager

# Load existing identity
identity = IdentityManager.load_from_file("my_agent.pem")

agent = AmorceAgent(
    llm=ChatOpenAI(),
    tools=[...],
    identity=identity  # Use specific identity
)

Custom Amorce Client

from amorce import AmorceClient

client = AmorceClient(
    identity,
    directory_url='https://my-directory.com',
    orchestrator_url='https://my-orchestrator.com'
)

agent = AmorceAgent(
    llm=ChatOpenAI(),
    tools=[...],
    amorce_client=client  # Custom client
)

---

🧪 Testing

# Run tests
pytest tests/

# With coverage
pytest --cov=langchain_amorce tests/

---

📚 Documentation

• Amorce Documentation
• LangChain Documentation
• A2A Protocol

---

🤝 Contributing

Contributions welcome! See CONTRIBUTING.md

---

📄 License

MIT License - see LICENSE

---

🔗 Links

• Amorce: amorce.io
• GitHub: github.com/amorce/langchain-amorce
• PyPI: pypi.org/project/langchain-amorce
• Issues: github.com/amorce/langchain-amorce/issues

---

Built with ❤️ by the Amorce team