Package io.quarkiverse.langchain4j.guardrails

Interface OutputGuardrail

@Experimental("This feature is experimental and the API is subject to change")
public interface OutputGuardrail

An output guardrail is a rule that is applied to the output of the model to ensure that the output is safe and meets the expectations.

Implementation should be exposed as a CDI bean, and the class name configured in OutputGuardrails.value() annotation.

Implementation should throw a OutputGuardrail.ValidationException when the validation fails. The exception can indicate whether the request should be retried and provide a reprompt message. In the case of reprompting, the reprompt message is added to the LLM context and the request is retried.

The maximum number of retries is configurable using quarkus.langchain4j.guardrails.max-retries, defaulting to 3.

Nested Class Summary

Nested Classes

Modifier and Type	Interface	Description
static final record	OutputGuardrail.OutputGuardrailParams	Represents the parameter passed to validate(OutputGuardrailParams).
static class	OutputGuardrail.ValidationException	An exception thrown when the validation fails.

Method Summary

Modifier and Type	Method	Description
default void	validate(dev.langchain4j.data.message.AiMessage responseFromLLM)	Validates the response from the LLM.
default void	validate(OutputGuardrail.OutputGuardrailParams params)	Validates the response from the LLM.

Method Details

validate

default void validate(dev.langchain4j.data.message.AiMessage responseFromLLM)
               throws OutputGuardrail.ValidationException

Validates the response from the LLM.

If the validation fails with an exception that is not a OutputGuardrail.ValidationException, no retry will be attempted.

Parameters:

responseFromLLM - the response from the LLM

Throws:

OutputGuardrail.ValidationException - the exception throws if the validation fails.

validate

default void validate(OutputGuardrail.OutputGuardrailParams params)
               throws OutputGuardrail.ValidationException

Validates the response from the LLM.

Unlike validate(AiMessage), this method allows to access the memory and the augmentation result (in the case of a RAG).

If the validation fails with an exception that is not a OutputGuardrail.ValidationException, no retry will be attempted.

Implementation must not attempt to write to the memory or the augmentation result.

Parameters:

params - the parameters, including the response from the LLM, the memory (maybe null), and the augmentation result (maybe null). Cannot be null

Throws:

OutputGuardrail.ValidationException - the exception throws if the validation fails.