Implementation:Evidentlyai Evidently Guard Decorator

Knowledge Sources	Evidentlyai_Evidently
Domains	Guardrails, Decorators, LLM_Safety
Last Updated	2026-02-14 12:00 GMT

Overview

guardrails/decorators.py provides the @guard decorator that wraps a function with guardrail validation, automatically intercepting a specified input argument and running it through one or more guardrails before the function executes.

Description

The module exposes a single decorator factory, guard(), which accepts:

• A single GuardrailBase instance or a list of them.
• An input_arg parameter name (default "input") indicating which function argument to validate.

When the decorated function is called, the decorator:

1. Uses inspect.signature to bind the positional and keyword arguments to parameter names.
2. Extracts the value of input_arg from the bound arguments.
3. If a list of guards is provided, calls validate_guards (from evidently.guardrails.core) to validate the input against all guards.
4. If a single guard is provided, calls guard.validate() directly.
5. If tracely is installed and a current tracing span exists, sets the guardrails as context values on the span for observability integration.
6. If validation passes, the original function is called normally.

This enables declarative, non-invasive guardrail enforcement on LLM pipeline functions.

Usage

Use the @guard decorator to protect functions (e.g., LLM call wrappers) with input validation guardrails. It is particularly useful in production pipelines where you want to ensure inputs meet safety or quality criteria before processing.

Code Reference

Source Location

• Repository: Evidentlyai_Evidently
• File: src/evidently/guardrails/decorators.py

Signature

def guard(
    guard: Union[GuardrailBase, List[GuardrailBase]],
    input_arg: str = "input",
) -> Callable:
    ...

Import

from evidently.guardrails.decorators import guard

I/O Contract

Inputs

Name	Type	Required	Description
guard	Union[GuardrailBase, List[GuardrailBase]]	Yes	A single guardrail or list of guardrails to validate against.
input_arg	str	No (default "input")	The name of the decorated function's parameter to validate.

Outputs

Name	Type	Description
return	Callable	A decorator that wraps the target function with guardrail validation.

Usage Examples

from evidently.guardrails.decorators import guard
from evidently.guardrails.core import GuardrailBase, GuardException

class ProfanityGuard(GuardrailBase):
    def validate(self, data: str):
        banned = ["badword"]
        for word in banned:
            if word in data.lower():
                raise GuardException(self, f"Profanity detected: {word}")

# Apply guard to a function, validating the 'prompt' argument
@guard(ProfanityGuard(), input_arg="prompt")
def call_llm(prompt: str) -> str:
    return "LLM response"

# Multiple guards
@guard([ProfanityGuard()], input_arg="prompt")
def call_llm_multi(prompt: str) -> str:
    return "LLM response"

Related Pages

• Environment:Evidentlyai_Evidently_Python_Core_Environment