Microsoft Foundry Hosted Agent Workshop

This repository contains a starter template for building an Microsoft Foundry that hosts AI agents using the Microsoft Foundry Hosted Agent Service.

Getting Started

Clone this repository to your local machine using the following command:

git clone https://github.com/hatasaki/hosted-agent-workshop.git
cd hosted-agent-workshop

Prerequisites

• Install azd
  • Windows: winget install microsoft.azd
  • Linux: curl -fsSL https://aka.ms/install-azd.sh | bash
  • MacOS: brew tap azure/azd && brew install azd

Quickstart

0. Create new directory and navigate into it:

   mkdir my-hosted-agent
   cd my-hosted-agent

1. Bring down the template code:

   azd init -t https://github.com/Azure-Samples/azd-ai-starter-basic

   • input your unique project name when prompted. Ex. <username>-hosted-agent

   This will perform a git clone

2. Sign into your Azure account:

   azd auth login

3. Download a sample agent definition:

   azd ai agent init -m ../msft-docs-agent/agent.yaml

• install agent extension if prompted.
• select your subscription when prompted.
• select supported region. (List of supported regions)
• select GlobalStandard for mode SKU.
• deployment name of model cab be left as default. (gpt-4o-mini)
• container memory, CPU, and replicas can be left as default. (2GB, 1 CPU, 1 min replica and 3 max replicas)

4. Provision Microsoft Foundry and the sample agent:

   azd up

5. Open Microsoft Foundry portal and test the agent

• Note: azd up creates new Microsoft Foundry account. Navigate to your Microsoft Foundry (should be listed under resource group rg-<username>-hosted-agent you specified in step 1).

Resource Clean-up

• Deleting Resources: To delete all associated resources and shut down the application, execute the following command:

  azd down

  Please note that this process may take up to 20 minutes to complete.

⚠️ Alternatively, you can delete the resource group directly from the Azure Portal to clean up resources.

Run the agent in a local environment

Prerequisites

• Install the Azure CLI

  • Windows: winget install --id Microsoft.AzureCLI
  • Linux: curl -sL https://aka.ms/InstallAzureCLIDeb | sudo bash
  • MacOS: brew update && brew install azure-cli

• You have already provisioned a Foundry project and deployed a model (for example by running azd up).

• Your signed-in user has Azure RBAC access to the Foundry project (at minimum the Azure AI User role on the project).

• Python 3.10+ is installed. Using a virtual environment (such as venv) is strongly recommended.

  • Example (Linux/MacOS):

    python3 -m venv .venv
    source .venv/bin/activate

  • Example (Windows PowerShell):

    python3 -m venv .venv
    .\.venv\Scripts\Activate.ps1

Steps

1. In a terminal, set the following environment variables using values from the Foundry portal (https://ai.azure.com):

   • AZURE_AI_PROJECT_ENDPOINT
   • AZURE_AI_MODEL_DEPLOYMENT_NAME

   Example (Linux/MacOS):

   export AZURE_AI_PROJECT_ENDPOINT="<your-project-endpoint>"
   export AZURE_AI_MODEL_DEPLOYMENT_NAME="<your-model-deployment-name>"

   Example (Windows PowerShell):

   $env:AZURE_AI_PROJECT_ENDPOINT = "<your-project-endpoint>"
   $env:AZURE_AI_MODEL_DEPLOYMENT_NAME = "<your-model-deployment-name>"

2. Sign in with Azure CLI and ensure you are using the tenant and subscription that contain your Foundry project:

   az login

   If needed, select a subscription:

   az account set --subscription "<subscription-id>"

3. Install dependencies:

   cd msft-docs-agent
   python3 -m pip install -r requirements.txt

4. Run the agent locally:

   python3 main.py

   This starts a local server (see sample.http for the request format).

5. Test the agent:

   • Using VS Code REST Client: open sample.http and send the request.
   • Or with curl:

     curl -sS http://0.0.0.0:8088/responses \
       -H 'Content-Type: application/json' \
       -d '{"input":"Explain overview of Microsoft Foundry in Japanese?"}'