Implementation:Guardrails ai Guardrails BaseFormatter

Knowledge Sources	Guardrails
Domains	Formatting, LLM Integration
Last Updated	2026-02-14 00:00 GMT

Overview

The BaseFormatter abstract class defines the interface for wrapping LLM callables with input/output manipulation logic, and PassthroughFormatter provides a no-op implementation.

Description

This module provides the formatter abstraction layer for Guardrails. A formatter takes an LLM callable and wraps it, enabling pre-processing of inputs or post-processing of outputs before they reach the validation pipeline. This is used for techniques such as JSON constrained-decoding.

Two classes are defined:

• BaseFormatter (ABC): Declares two abstract methods:
  • wrap_callable: Wraps a synchronous PromptCallableBase into an ArbitraryCallable.
  • wrap_async_callable: Wraps a synchronous PromptCallableBase into an AsyncPromptCallableBase for async execution contexts.

• PassthroughFormatter: A concrete implementation that returns the original callable unchanged (no-op). This is the default formatter when no special input/output manipulation is needed.

Usage

Use BaseFormatter as the base class when implementing custom formatters that need to modify LLM inputs or outputs (e.g., constraining the LLM to produce valid JSON). Use PassthroughFormatter as the default when no formatting is required.

Code Reference

Source Location

• Repository: Guardrails
• File: guardrails/formatters/base_formatter.py
• Lines: 1-32

Signature

class BaseFormatter(ABC):
    @abstractmethod
    def wrap_callable(self, llm_callable: PromptCallableBase) -> ArbitraryCallable: ...

    @abstractmethod
    def wrap_async_callable(
        self, llm_callable: PromptCallableBase
    ) -> AsyncPromptCallableBase: ...


class PassthroughFormatter(BaseFormatter):
    def wrap_callable(self, llm_callable: PromptCallableBase):
        return llm_callable  # Noop

    def wrap_async_callable(self, llm_callable: PromptCallableBase):
        return llm_callable  # Noop

Import

from guardrails.formatters.base_formatter import BaseFormatter, PassthroughFormatter

I/O Contract

wrap_callable

Parameter	Type	Description
llm_callable	PromptCallableBase	The synchronous LLM callable to wrap.

Return Type	Description
ArbitraryCallable	The wrapped callable with any input/output transformations applied.

wrap_async_callable

Parameter	Type	Description
llm_callable	PromptCallableBase	The LLM callable to wrap for async execution.

Return Type	Description
AsyncPromptCallableBase	The wrapped async callable with any input/output transformations applied.

PassthroughFormatter Behavior

Method	Behavior
wrap_callable	Returns the input llm_callable unchanged.
wrap_async_callable	Returns the input llm_callable unchanged.

Usage Examples

from guardrails.formatters.base_formatter import BaseFormatter, PassthroughFormatter
from guardrails.llm_providers import PromptCallableBase, ArbitraryCallable

# Using the PassthroughFormatter (no-op)
formatter = PassthroughFormatter()
original_callable = my_llm_callable  # some PromptCallableBase instance
wrapped = formatter.wrap_callable(original_callable)
# wrapped is the same object as original_callable

# Implementing a custom formatter
class JsonConstrainedFormatter(BaseFormatter):
    def wrap_callable(self, llm_callable: PromptCallableBase) -> ArbitraryCallable:
        # Add JSON constrained decoding logic
        def wrapped_fn(*args, **kwargs):
            result = llm_callable(*args, **kwargs)
            # Post-process to ensure valid JSON
            return result
        return ArbitraryCallable(wrapped_fn)

    def wrap_async_callable(self, llm_callable: PromptCallableBase):
        # Async version of the wrapper
        async def wrapped_fn(*args, **kwargs):
            result = await llm_callable(*args, **kwargs)
            return result
        return wrapped_fn

Related Pages

• Implementation:Guardrails_ai_Guardrails_OutputTypes - Output type enumeration that formatters may use to determine output structure
• Implementation:Guardrails_ai_Guardrails_Filter_Action - Post-validation action that may interact with formatted outputs
• Implementation:Guardrails_ai_Guardrails_Refrain_Action - Post-validation action that may interact with formatted outputs