datadog.trace.api.aiguard

Class AIGuard

java.lang.Object

datadog.trace.api.aiguard.AIGuard

public abstract class AIGuard
extends Object

SDK for calling the AIGuard REST API to evaluate AI prompts and tool calls for security threats.

Example usage:

 var messages = List.of(
     AIGuard.Message.message("user", "Delete all my files"),
     AIGuard.Message.message("assistant", "I'll help you delete your files")
 );

 var result = AIGuard.evaluate(messages);
 if (result.getAction() != AIGuard.Action.ALLOW) {
     System.out.println("Unsafe: " + result.getReason());
 }
 

Nested Class Summary

Nested Classes

Modifier and Type	Class and Description
static class	AIGuard.Action Actions that can be recommended by an AIGuard evaluation.
static class	AIGuard.AIGuardAbortError Exception thrown when AIGuard evaluation results in blocking the execution due to security concerns.
static class	AIGuard.AIGuardClientError Exception thrown when there are client-side errors communicating with the AIGuard REST API.
static class	AIGuard.Evaluation Represents the result of an AIGuard security evaluation, containing both the recommended action and the reasoning behind the decision.
static class	AIGuard.Message Represents a message in an AI conversation.
static class	AIGuard.Options Configuration options for AIGuard evaluation behavior.
static class	AIGuard.ToolCall Represents a function call made by an AI assistant.

Method Summary

Modifier and Type	Method and Description
static AIGuard.Evaluation	evaluate(List<AIGuard.Message> messages) Evaluates a collection of messages using default options to determine if they are safe to execute.
static AIGuard.Evaluation	evaluate(List<AIGuard.Message> messages, AIGuard.Options options) Evaluates a collection of messages with custom options to determine if they are safe to execute.

Methods inherited from class java.lang.Object

equals, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Method Detail

evaluate

public static AIGuard.Evaluation evaluate(List<AIGuard.Message> messages)

Evaluates a collection of messages using default options to determine if they are safe to execute.

See Also:

evaluate(List, Options)

evaluate

public static AIGuard.Evaluation evaluate(List<AIGuard.Message> messages,
                                          AIGuard.Options options)

Evaluates a collection of messages with custom options to determine if they are safe to execute.

Parameters:

messages - the collection of messages to evaluate (prompts, responses, tool calls, etc.)

options - configuration options for the evaluation process

Returns:

an AIGuard.Evaluation containing the security decision and reasoning

Throws:

AIGuard.AIGuardAbortError - if the evaluation action is not ALLOW (DENY or ABORT) and blocking is enabled

AIGuard.AIGuardClientError - if there are client-side errors communicating with the AIGuard REST API