langchain - 💡(How to fix) Fix RFC: ComplianceCallbackHandler — tamper-evident audit trails for regulated industries [13 comments, 5 participants]

ComplianceCallbackHandler Pattern for LangChain

Fix Plan

To implement the ComplianceCallbackHandler pattern for generating tamper-evident audit trails, follow these steps:

1. Install AgentMint: Clone the AgentMint repository and install it using pip:

git clone https://github.com/aniketh-maddipati/agentmint-python.git cd agentmint-python pip install .

2. **Create a ComplianceCallbackHandler class**: In your LangChain project, create a new class that inherits from `BaseCallbackHandler` and uses the `AgentMintClient` to generate receipts:
   ```python
from langchain_core.callbacks import BaseCallbackHandler
from agentmint import AgentMintClient

class ComplianceCallbackHandler(BaseCallbackHandler):
    def __init__(self):
        self.mint = AgentMintClient(policy="policy.yaml")

    def on_tool_end(self, output, *, run_id, **kwargs):
        self.mint.notarize(run_id=str(run_id), output=output)

1. Register the ComplianceCallbackHandler: In your LangChain configuration, register the ComplianceCallbackHandler as a callback handler:

from langchain_core import LangChain

langchain = LangChain(callback_handlers=[ComplianceCallbackHandler()])

4. **Use the ComplianceCallbackHandler**: When using LangChain, make sure to use the `ComplianceCallbackHandler` as the callback handler:
   ```python
langchain.run_tool("your_tool_name", callback_handler=ComplianceCallbackHandler())

Verification

To verify that the fix worked, check the following:

• The ComplianceCallbackHandler is registered correctly in the LangChain configuration.
• The AgentMintClient is generating receipts correctly for each tool call.
• The receipts are cryptographically signed and timestamped.
• The receipts can be verified independently using OpenSSL.

Extra Tips

• Make sure to update