[bug] Redact Call.inputs when the class is printed/logged

[CalebCourier]

Describe the bug
Sometimes users need to pass sensitive information as kwargs when running a Guard. One example is api_key. When a Guard is called, i.e. guard(), kwargs are captured as part of the Call class for Guard History. The history objects stay on the client machine, but should still be redacted when printed or logged by the client.

To Reproduce

import os
from guardrails import Guard
from rich import print

guard = Guard()


guard(
    model="gpt-3.5-turbo",
    messages=[{
        "role": "user",
        "content": "Hello, world!"
    }],
    api_key=os.environ.get("OPENAI_API_KEY")
)


print(guard.history.last.inputs)

Expected behavior
Potentially sensitive kwargs should be redacted.

Library version:
0.6.x

Additional context
We already have redaction logic in Call.to_interface for this exact reason. We should just be able to override __str__ and __repr__ and reuse the same redaction technique.
See also:
#1213
#1217

[abhishek9sharma]

Hi @CalebCourier I was trying something like this abhishek9sharma@3a9513d.
It works for a normal print but not when I use rich. It does work for both when I use print(str(guard.history.last.inputs))
not sure what I am missing something here.

[abhishek9sharma]

Any guidance on this ?

[abhishek9sharma]

@CalebCourier any advice on this

[CalebCourier]

Hi @abhishek9sharma, sorry for the delayed response. I think it makes sense that __str__ is working only for print. Try also adding __repr__ for rich.