[FEATURE] Support remote guardrails model

[jngz-es]

Is your feature request related to a problem?
Improve guardrails performance.

What solution would you like?
Call remote guardrails model to check user input and model output.

What alternatives have you considered?
We already have an alternative using stop words and regex to validate input/output.

Do you have any additional context?
None.

[xinyual]

Do we have any design for that? For example, we will call it parallel when we using agent execute API? Can we design it as a function/method so I can reuse it in PPL Tool in skills repo?

[jngz-es]

Hi @xinyual , we plan to only add a new type for the current guardrail implementation. I believe you can reuse it for PPL tool, as it works on model level, runs in model prediction process. So it doesn't matter with agent execute API.