Develop an AI app with the Microsoft Foundry SDK

Develop an AI app with the Microsoft Foundry SDK

• Microsoft Foundry provides a REST API to work with:

  • AI Foundry projects

  • Resources inside those projects

• It also offers language-specific SDKs to make coding easier.

• Using the SDK, developers can:

  • Connect to a Foundry project

  • Access project resources and models

  • Perform AI operations (e.g., send prompts to generative AI models and process responses)

Core SDK Library

• The main library is Azure AI Projects

• It allows you to:

  • Connect to a Microsoft Foundry project

  • Access all resources defined in the project

Available SDKs:

• Azure AI Projects for Python

• Azure AI Projects for .NET

• Azure AI Projects for JavaScript

Each SDK is developed independently, so features may differ slightly across languages.

Language Used in This Module

• This module uses Python examples

• Equivalent examples are available in other SDK documentation

Install the Azure AI SDK :

• pip install azure-ai-projects

Connecting to a Microsoft Foundry Project

Project Endpoint

• Each project has a unique endpoint

• You can find it on the Project Overview page in the Microsoft Foundry portal

Types of Endpoints in a Project:

1. Project endpoint

   • Access project connections, agents, and models

2. Azure OpenAI Service endpoint

   • Use OpenAI models

3. Foundry Tools endpoint

   • Use tools like Azure Vision and Azure Language

Creating a project Client

• Use the project endpoint to create a project client

from azure.identity import DefaultAzureCredential

from azure.ai.projects import AIProjectClient

project_endpoint = "https://......"

project_client = AIProjectClient(

    credential=DefaultAzureCredential(),

    endpoint=project_endpoint

)

Authentication

• Uses DefaultAzureCredential

• You must install the identity package:

• pip install azure-identity

Important Authentication Requirement

• Code must run in an authenticated Azure session

• Example:

  • Run az login using Azure CLI before executing the code

Summary

• Microsoft Foundry SDK lets developers programmatically work with AI projects

• Azure AI Projects is the core library

• Authentication and the correct project endpoint are mandatory

Project Connections

• Every Microsoft Foundry project includes connected resources

• These resources are defined at:

  • Parent level (Microsoft Foundry resource / hub)

  • Project level

• Each connection links to an external service, such as:

  • Azure Storage

  • Azure AI Search

  • Azure OpenAI

  • Another Microsoft Foundry resource

Why Project Connections Matter

• Connections allow your project to use external services

• Using the Microsoft Foundry SDK, you can:

  • Connect to a project

  • Retrieve its connections

  • Use those connections to consume services programmatically

Accessing Connections Using the SDK

• In Python, the AIProjectClient object has a connections property

• This property lets you list and retrieve resource connections

Important Connection Methods

connections.list()

• Returns all connection objects in the project

• Each object represents one connected resource

• You can filter results using:

  • connection_type parameter

  • Example: ConnectionType.AZURE_OPEN_AI

connections.get(connection_name, include_credentials)

• Returns a specific connection by name

• include_credentials:

  • Default: True

  • If True, credentials are included (e.g., API keys for Foundry Tools)

What Connection Objects Contain

• Connection name

• Connection type

• Connection-specific properties

• Credentials (if included)

• These details are used to connect to the external resource

Python Example: List All Project Connections

from azure.identity import DefaultAzureCredential
from azure.ai.projects import AIProjectClient

try:
    # Get project client
    project_endpoint = "https://....."
    project_client = AIProjectClient(
        credential=DefaultAzureCredential(),
        endpoint=project_endpoint,
    )

    # List all connections in the project
    connections = project_client.connections
    print("List all connections:")
    for connection in connections.list():
        print(f"{connection.name} ({connection.type})")

except Exception as ex:
    print(ex)

Summary

• Project connections link Foundry projects to external Azure services

• Connections can be accessed using AIProjectClient.connections

• You can:

  • List all connections

  • Get a specific connection

  • Retrieve credentials if needed

Purpose

• A common AI app scenario is to:

  • Connect to a generative AI model

  • Use prompts to have a chat-based conversation

Two Ways to Connect to Models

1. Directly using Azure OpenAI SDK

   • Authentication via:

     • API keys

     • Microsoft Entra ID

2. Using Microsoft Foundry SDK (Recommended for Foundry projects)

   • Retrieve a project client

   • From it, get an authenticated OpenAI chat client

   • Works for any model deployed in the project

Why Use Microsoft Foundry SDK?

• Simplifies consuming models deployed in your project

• Easy to switch between models

  • Just change the model deployment name

• Automatically handles authentication

Important Tip

• The OpenAI chat client from Microsoft Foundry can chat with:

  • Azure OpenAI models

  • Non-OpenAI models (e.g., Microsoft Phi models)
    as long as they are deployed in the Foundry resource

Required Python Packages

pip install azure-ai-projects
pip install azure-identity
pip install openai

Steps to Create a Chat Client

1. Authenticate using DefaultAzureCredential

2. Connect to the Microsoft Foundry project

3. Get an OpenAI chat client using get_openai_client()

4. Send a prompt and receive a chat completion

Python Example: Create a Chat Client

from azure.identity import DefaultAzureCredential
from azure.ai.projects import AIProjectClient
from openai import AzureOpenAI

try:
    # Connect to the project
    project_endpoint = "https://......"
    project_client = AIProjectClient(
        credential=DefaultAzureCredential(),
        endpoint=project_endpoint,
    )

    # Get an OpenAI chat client
    chat_client = project_client.get_openai_client(
        api_version="2024-10-21"
    )

    # Get user input
    user_prompt = input("Enter a question:")

    # Create chat completion
    response = chat_client.chat.completions.create(
        model=your_model_deployment_name,
        messages=[
            {"role": "system", "content": "You are a helpful AI assistant."},
            {"role": "user", "content": user_prompt}
        ]
    )

    print(response.choices[0].message.content)

except Exception as ex:
    print(ex)

Key Components

• project_endpoint → Foundry project endpoint

• get_openai_client() → Returns authenticated chat client

• model → Deployment name of the model in Foundry

• messages → Conversation history (system + user roles)

Summary

• Microsoft Foundry SDK provides an easy way to chat with deployed models

• Supports multiple models, including non-OpenAI models

• Authentication is handled using Azure credentials

• Ideal for building chat-based generative AI apps