Adding Audit Trails to LangChain Agents with Asqav

#python #security #ai #tutorial

LangChain agents make decisions autonomously. They call tools, query APIs, and process data. But once they run in production, you need to know exactly what they did.

The gap

LangChain has callbacks for logging, but no built-in:

Cryptographic proof of what happened
Policy enforcement to prevent risky actions
Compliance-ready audit reports

Adding governance

pip install asqav

Basic integration

from asqav import Asqav

client = Asqav(api_key="sk_...")

# Create an agent identity
agent = client.create_agent(
    name="langchain-research-agent",
    algorithm="ML-DSA-65"
)

# Before each tool call, sign the action
def governed_tool_call(tool_name, tool_input):
    # Check policy first
    sig = client.sign(
        agent_id=agent.agent_id,
        action_type=f"tool:{tool_name}",
        action_id=f"{tool_name}-{id(tool_input)}",
        payload=tool_input
    )
    print(f"Audit record: {sig.signature_id}")
    # Proceed with the actual tool call
    return tool.invoke(tool_input)

Policy enforcement

Block specific actions before they execute:

# Create a policy via the API
client.create_policy(
    name="no-external-apis",
    action_pattern="tool:requests_*",
    action="block_and_alert",
    severity="high"
)

Now any LangChain tool call matching requests_* gets blocked.

What you get

Every tool call in your LangChain agent gets:

A quantum-safe cryptographic signature (ML-DSA)
Timestamp, action type, and payload recorded
Policy check before execution
Verifiable audit trail for compliance

Dashboard

All signatures appear in the Asqav dashboard where you can:

Search and filter by agent, action type, or time range
Generate compliance reports
Set up alerts for policy violations

DEV Community