Azure AI Agents SDK for Java

Develop agents using Azure AI Foundry with an extensive ecosystem of models, tools, and capabilities.

Installation

<dependency>
    <groupId>com.azure</groupId>
    <artifactId>azure-ai-agents</artifactId>
    <version>1.0.0-beta.1</version>
</dependency>

Environment Variables

PROJECT_ENDPOINT=https://<resource>.services.ai.azure.com/api/projects/<project>
MODEL_DEPLOYMENT_NAME=gpt-4o-mini

Authentication

import com.azure.ai.agents.AgentsClient;
import com.azure.ai.agents.AgentsClientBuilder;
import com.azure.identity.DefaultAzureCredentialBuilder;

String endpoint = System.getenv("PROJECT_ENDPOINT");
AgentsClient agentsClient = new AgentsClientBuilder()
    .endpoint(endpoint)
    .credential(new DefaultAzureCredentialBuilder().build())
    .buildClient();

Client Hierarchy

The SDK provides three main sub-clients:

AgentsClientBuilder builder = new AgentsClientBuilder()
    .credential(new DefaultAzureCredentialBuilder().build())
    .endpoint(endpoint);

// Build sub-clients
AgentsClient agentsClient = builder.buildClient();
ConversationsClient conversationsClient = builder.buildConversationsClient();
ResponsesClient responsesClient = builder.buildResponsesClient();

Core Workflow

1. Create a Prompt Agent

import com.azure.ai.agents.models.PromptAgentDefinition;
import com.azure.ai.agents.models.AgentVersionDetails;

PromptAgentDefinition definition = new PromptAgentDefinition("gpt-4o");
AgentVersionDetails agent = agentsClient.createAgentVersion("my-agent", definition);

2. Create Conversation

Conversation conversation = conversationsClient.getConversationService().create();

3. Add Messages to Conversation

import com.openai.models.EasyInputMessage;
import com.openai.models.ItemCreateParams;

conversationsClient.getConversationService().items().create(
    ItemCreateParams.builder()
        .conversationId(conversation.id())
        .addItem(EasyInputMessage.builder()
            .role(EasyInputMessage.Role.SYSTEM)
            .content("You are a helpful assistant.")
            .build())
        .addItem(EasyInputMessage.builder()
            .role(EasyInputMessage.Role.USER)
            .content("Hello, agent!")
            .build())
        .build()
);

4. Generate Response

import com.azure.ai.agents.models.AgentReference;

AgentReference agentRef = new AgentReference(agent.getName())
    .setVersion(agent.getVersion());

Response response = responsesClient.createWithAgentConversation(
    agentRef,
    conversation.id()
);

Using OpenAI’s Official Library

The SDK transitively imports the OpenAI Java SDK:

// Access OpenAI services directly
ResponsesService responsesService = responsesClient.getOpenAIClient();
ConversationService conversationService = conversationsClient.getOpenAIClient();

Best Practices

1. Use DefaultAzureCredential for production authentication
2. Share conversations across multiple agents for centralized context
3. Use async clients for better throughput
4. Clean up conversations when no longer needed
5. Handle errors with appropriate exception handling

Error Handling

import com.azure.core.exception.HttpResponseException;

try {
    AgentVersionDetails agent = agentsClient.createAgentVersion(name, definition);
} catch (HttpResponseException e) {
    System.err.println("Error: " + e.getResponse().getStatusCode());
}

Reference Links