Default On Guardrails and output

[pierrelouisbess]

Hi !

I have 2 question about guardrails :
1)
I'm new to Litellm and the guardrails generated by policy templates, and I'd like to know if it's possible to automatically set them all to “default on”?
Currently, I have to manually click on each guardrail, edit it, and set it to “default on.”

I've been testing pre-call guardrails, and I'd like to know if it's possible to see the output sent to the LLM after my guardrail's (Mask) action. Currently, I can only see the user's unmasked request and the LLM's response, but not the prompt that has been modified into output by the guardrails.

Thanks !

[jagmarques]

For (1), you can set guardrails to default-on via the config YAML rather than the UI. Under your litellm_settings, add a guardrails list with default_on: true for each one. That way they apply to all requests without manual toggling per guardrail.

For (2), the masked/modified prompt sent to the LLM is not currently exposed in the standard logs UI. A workaround is to add a custom callback (litellm.callbacks) that captures the kwargs["messages"] in the async_pre_call_hook - at that point the guardrail masking has already been applied, so you can log the actual prompt that gets sent downstream.