AgentTrust Docs

Overview

AgentTrust integrates with LangChain via auto_instrument() (patches LCEL and legacy chains) and @harness on chain invocation wrappers. No dedicated adapter module — integration uses core SDK hooks.

Tier: OSS+ (auto-instrument) · Team for advanced graph node patterns

Why It Matters

LangChain is one of the most common agent frameworks. Governance must work with LCEL pipelines, tool calls, and chain callbacks without rewriting chain definitions.

Prerequisites

pip install agentrust-sdk langchain langchain-openai

Step-by-Step Guide

Pattern A — Auto-instrument

from agentrust_sdk import auto_instrument, embed_gateway
embed_gateway()
auto_instrument()

from langchain_openai import ChatOpenAI
from langchain_core.prompts import ChatPromptTemplate

prompt = ChatPromptTemplate.from_template("Answer: {input}")
chain = prompt | ChatOpenAI(model="gpt-4o-mini")
result = chain.invoke({"input": "What is AI governance?"})

Pattern B — `@harness` wrapper

from agentrust_sdk import harness

@harness(agent_id="langchain-agent", framework="LangChain")
def run_chain(user: str, input: str) -> dict:
    result = chain.invoke({"input": input})
    return {"answer": result.content}

Pattern C — Direct client

from agentrust_sdk import AgentTrustClient

with AgentTrustClient() as client:
    output = chain.invoke({"input": "..."})
    result = client.validate(
        agent_id="langchain-agent",
        user="alice",
        input="...",
        output={"answer": output.content},
        framework="LangChain",
    )

Examples

See examples/langchain_agent.py for all three patterns.

Best Practices

Call auto_instrument() before LangChain imports
Include tool call metadata in validation for tool trust checks
Use @harness with explicit agent_id in production
For LangGraph graphs, use the LangGraph adapter instead

Common Mistakes

Confusing LangChain hooks with LangGraph AgentTrustNode (different adapters)
Not returning structured dict from @harness wrapper
Import order: auto_instrument must precede langchain imports

Troubleshooting

Issue	Fix
Chain calls not validated	Check import order and `AGENTRUST_AUTO_INSTRUMENT`
Tool trust failures	Pass `tool_calls` in direct client validation
LangGraph in same app	Use LangGraph adapter for graph nodes

LangChain Integration