Custom guardrails for the agent

[vnalawad-tibco]

I see callbacks for agent/LLM/Tool but no way to specify custom request/response processors for implementing guardrails.
e.g. sensitive data redaction before sending request/response to LLM for processing.
How this can be implemented today?

[dpasiukevich]

It's possible to do it in before_model_callback. The callback is run before the request is sent to LLM and it allows to update the request before it's sent.