Agentic Azure App Service app with Semantic Kernel and Foundry Agent Service (Spring WebFlux)

This is a Spring Boot WebFlux application that demonstrates AI agent integration using both Semantic Kernel and Foundry Agent Service for intelligent task management. It provides a simple CRUD task list and two interactive chat agents.

Getting Started

See Tutorial: Build an agentic web app in Azure App Service with Microsoft Semantic Kernel or Foundry Agent Service (Spring Boot).

Features

• CRUD Operations: Create, Read, Update, Delete tasks using REST API.
• OpenAPI Documentation: Auto-generated OpenAPI schema at /api/schema for Foundry agent integration.
• Semantic Kernel Agent: Local agent orchestration with automatic function calling.
• Foundry Agent: Cloud-hosted agent with managed function calling.
• Dual Agent UI: Compare and interact with both agent frameworks side-by-side.
• Azure Deployment: Azure Developer CLI (azd) template can create App Service with managed identity using azd up.

Implementation Notes

Semantic Kernel Agent

The Semantic Kernel agent implementation uses the ChatCompletionAgent with automatic function calling:

1. ChatCompletionAgent: Uses Microsoft's Semantic Kernel ChatCompletionAgent for intelligent task management.
2. Automatic Function Calling: Configured with FunctionChoiceBehavior.auto(true) to automatically call appropriate plugin functions.
3. Direct Plugin Integration: Uses @DefineKernelFunction annotations in TaskCrudPlugin for seamless integration.
4. Local Orchestration: Agent runs in the application with direct access to task database.

Foundry Agent

The Foundry Agent implementation uses the Azure AI Agents SDK with async clients:

1. AgentsAsyncClient: Uses Azure's cloud-hosted agent service for intelligent task management.
2. Managed Function Calling: Agent and tools configured in the Foundry portal using the OpenAPI schema.
3. OpenAPI Integration: REST API endpoints are documented with OpenAPI annotations and exposed at /api/schema.
4. Conversation Management: Each session maintains its own conversation thread.
5. Cloud Orchestration: Agent runs in Azure with API-based access to task functions.

Project Structure

├── .devcontainer/
│   └── devcontainer.json                          # Codespace definition
├── azure.yaml                                     # Azure Developer CLI configuration
├── pom.xml                                        # Maven project configuration
├── infra/                                         # Azure Developer CLI configuration
│   ├── main.bicep
│   └── main.parameters.json
└── src/
    └── main/
        ├── java/
        │   └── com/example/crudtaskswithagent/
        │       ├── CrudTasksWithAgentApplication.java  # Main application class
        │       ├── config/
        │       │   └── OpenApiConfig.java              # OpenAPI/Swagger configuration
        │       ├── controller/
        │       │   ├── TaskController.java             # REST API endpoints
        │       │   └── AgentController.java            # AI agent endpoints (SK & Foundry)
        │       ├── model/
        │       │   └── TaskItem.java                   # Task entity
        │       ├── plugin/
        │       │   └── TaskCrudPlugin.java             # Semantic Kernel plugin
        │       ├── repository/
        │       │   └── TaskRepository.java             # Data repository
        │       └── service/
        │           ├── SemanticKernelAgentService.java # SK agent service
        │           └── FoundryAgentService.java        # Foundry Agent Service
        └── resources/
            ├── application.properties                  # Application configuration
            ├── schema.sql                              # Database schema
            └── static/
                └── index.html                          # Frontend UI (dual agent)

Configuration

The application requires configuration for both agent types:

OpenAPI Documentation

The API schema is auto-generated and available at:

springdoc.api-docs.path=/api/schema

Access the schema at http://localhost:8080/api/schema when running locally, or https://your-app.azurewebsites.net/api/schema when deployed.

Semantic Kernel (Azure OpenAI)

azure.openai.endpoint=https://your-openai-resource.openai.azure.com/
azure.openai.deployment=gpt-4o

Foundry Agent

azure.foundry.endpoint=https://<resource-name>.services.ai.azure.com/api/projects/<project-name>
azure.foundry.agent.name=your-agent-name

Both agents use Azure managed identity for authentication.