langchain-agentoath

AgentOath trust protocol integration for LangChain.

Automatically generates cryptographically signed trust receipts for every LangChain chain execution, tool call, and LLM interaction.

Installation

pip install langchain-agentoath

For full LangChain integration (optional):

pip install langchain-agentoath[langchain]

Quick Start

Callback Handler -- automatic receipts

from agentoath import TrustAgent
from langchain_agentoath import AgentOathCallbackHandler

# Create an agent identity
agent = TrustAgent.create(name="MyChain", capabilities=["chat"])

# Create the callback handler
handler = AgentOathCallbackHandler(agent)

# Attach to any LangChain chain
result = chain.invoke(
    {"input": "What is the weather?"},
    config={"callbacks": [handler]},
)

# Every chain/tool/LLM event now has a signed receipt
for receipt in handler.receipts:
    print(receipt.action, receipt.rating, receipt.description)

TrustedChain -- wrapper with pre/post receipts

from langchain_agentoath import TrustedChain

trusted = TrustedChain(chain=my_chain, agent=agent)
result = trusted.invoke({"input": "hello"})

# Pre-execution + post-execution + per-step receipts
print(f"Generated {len(trusted.receipts)} trust receipts")

Trust-aware Tools

from langchain_agentoath import VerifyAgentTool, SignReceiptTool, CheckTrustTool

tools = [
    VerifyAgentTool(trust_agent=agent),
    SignReceiptTool(trust_agent=agent),
    CheckTrustTool(trust_agent=agent, min_score=0.7),
]

# Use in any LangChain agent
result = tools[0].run('{"agent_did": "did:trust:agent:abc123"}')

TrustFilteredRetriever -- trust-gated RAG

from langchain_agentoath import TrustFilteredRetriever

retriever = TrustFilteredRetriever(
    base_retriever=my_vector_store.as_retriever(),
    agent=agent,
    min_trust_score=0.6,
)

# Only returns documents from agents with trust score >= 0.6
docs = retriever.get_relevant_documents("important query")

Components

License

Apache-2.0