OutputGuardrail¶

This docs was updated at: 2026-02-23

com.paragon.agents.OutputGuardrail · Interface

Validates agent output before returning to the user.

Output guardrails are executed after the agent has finished processing, just before the final result is returned. They can be used to:

Filter or mask sensitive information in responses
Enforce output length limits
Check for inappropriate content
Validate structured output format

Usage Example¶

OutputGuardrail maxLength = (output, ctx) -> {
    if (output.length() > 1000) {
        return GuardrailResult.failed("Response exceeds 1000 characters");
    }
    return GuardrailResult.passed();
};
Agent agent = Agent.builder()
    .addOutputGuardrail(maxLength)
    .build();

See Also

InputGuardrail
GuardrailResult

Since: 1.0

Methods¶

`validate`¶

GuardrailResult validate(@NonNull String output, @NonNull AgenticContext context)

Validates the agent's output.

Parameters

Name	Description
`output`	the agent's output string
`context`	the current agent context

Returns

the validation result

`named`¶

static @NonNull OutputGuardrail named(@NonNull String id, @NonNull OutputGuardrail impl)

Wraps a guardrail implementation with a named ID for blueprint serialization.

Use this when defining guardrails as lambdas that need to be serializable. The guardrail is registered in the GuardrailRegistry and can be reconstructed during deserialization.

OutputGuardrail guard = OutputGuardrail.named("max_length", (output, ctx) -> {
    if (output.length() > 1000) return GuardrailResult.failed("Too long");
    return GuardrailResult.passed();
});