Add Guardrails as a tool

[bboynton97]

As an MVP, give agents the ability to validate output against guardrails:

pii_guard.validate("My phone number is 555-555-6868")

Example code from a meeting:

from openai import OpenAI

from guardrails import Guard, install

# CLI Installation
# guadrails hub install hub://guadrails/detect_pii

install("hub://guardrails/detect_pii")
install("hub://guardrails/toxic_language")
from guardrails.hub import DetectPII
from guardrails.hub import ToxicLanguage

client = OpenAI()
pii_guard = Guard().use(DetectPII(pii_entities=["EMAIL_ADDRESS", "PHONE_NUMBER"]))
toxic_language_guard = Guard().use(ToxicLanguage())
combo_guard = Guard().use_many(DetectPII(pii_entities=["EMAIL_ADDRESS", "PHONE_NUMBER"]), ToxicLanguage())

client(
    model='gpt-4o-mini',
    messages=[{
        "role": "user",
        "content": "What is the meaning of life?"
    }]
)

pii_guard(
    model='gpt-4o-mini',
    messages=[{
        "role": "user",
        "content": "What is the meaning of life?"
    }]
)

toxic_language_guard(
    model='gpt-4o-mini',
    messages=[{
        "role": "user",
        "content": "What is the meaning of life?"
    }]
)

pii_guard.validate("My phone number is 555-555-6868")