MAF-MultiAgents

MAF-MultiAgents - Multi-Model Orchestration Demo

This sample demonstrates how to orchestrate multiple AI agents using different AI providers in a sequential workflow with the Microsoft Agent Framework.

Objective

This demo showcases a real-world scenario where three specialized agents collaborate to research, write, and review an article:

1. Researcher Agent (Microsoft Foundry Persistent Agent) - Researches topics and gathers key information
2. Writer Agent (Azure OpenAI) - Creates engaging content based on research
3. Reviewer Agent (Ollama local model) - Reviews the article and provides constructive feedback

The workflow demonstrates:

• Multi-provider orchestration: Using different AI services (Microsoft Foundry, Azure OpenAI, and Ollama) in a single workflow
• Sequential agent workflows: Agents execute in order, with each building on the previous agent's output
• Persistent agents: Creating and managing agents in Microsoft Foundry
• OpenTelemetry tracing: Monitoring agent execution with distributed tracing
• Flexible configuration: Supporting multiple authentication methods and providers

Architecture

User Input
    ↓
Researcher (Microsoft Foundry Agent)
    ↓ (research findings)
Writer (Azure OpenAI)
    ↓ (article draft)
Reviewer (Ollama - llama3.2)
    ↓
Final Output (reviewed article with feedback)

Prerequisites

Before running this sample, ensure you have:

1. .NET 10 SDK or later installed
2. Azure CLI installed and authenticated (az login)
3. Ollama installed and running locally with the llama3.2 model
4. Microsoft Foundry Project with access to deploy models
5. One of the following for Agent 2 (Writer):
   • Azure OpenAI endpoint and API key
   • Azure OpenAI endpoint with managed identity/Azure CLI credentials

Setup Instructions

Step 1: Install and Configure Ollama

Ollama is required for the Reviewer agent (Agent 3).

# Install Ollama from https://ollama.com

# Pull the llama3.2 model
ollama pull llama3.2

# Verify Ollama is running (should respond on http://localhost:11434/)
ollama run llama3.2

Step 2: Configure User Secrets

This application uses .NET user secrets to store sensitive configuration. Navigate to the project directory and set the required secrets:

cd 06-MAF/src/MAF-MultiAgents

Required: Microsoft Foundry Configuration (Agent 1 - Researcher)

# Set your Microsoft Foundry project endpoint
dotnet user-secrets set "AZURE_FOUNDRY_PROJECT_ENDPOINT" "https://<your-project>.services.ai.azure.com/"

# Set your model deployment name (default: gpt-5-mini)
dotnet user-secrets set "deploymentName" "gpt-5-mini"

Required: Writer Agent Configuration (Agent 2)

Choose ONE of the following options:

dotnet user-secrets set "deploymentName" "gpt-5-mini"

To create a GitHub token:

1. Go to https://github.com/settings/tokens
2. Click "Generate new token (classic)"
3. Select scopes as needed
4. Copy the token and use it above

Option B: Azure OpenAI with API Key

dotnet user-secrets set "AzureOpenAI:Endpoint" "https://<your-resource>.openai.azure.com/"
dotnet user-secrets set "AzureOpenAI:ApiKey" "your-azure-openai-api-key"
dotnet user-secrets set "AzureOpenAI:Deployment" "gpt-5-mini"

Option C: Azure OpenAI with DefaultAzureCredential (Recommended)

dotnet user-secrets set "AzureOpenAI:Endpoint" "https://<your-resource>.openai.azure.com/"
dotnet user-secrets set "AzureOpenAI:Deployment" "gpt-5-mini"
az login  # Required for DefaultAzureCredential

Step 3: Verify Azure CLI Authentication

The Researcher agent uses Azure CLI credentials to authenticate with Microsoft Foundry:

# Login to Azure CLI
az login

# Verify your authentication
az account show

Configuration Priority

The application automatically selects the chat client provider in the following order:

2. Azure OpenAI with API Key - If apikey is set
3. Azure OpenAI with Default Credentials - Fallback using Azure CLI or managed identity

Running the Application

Once all secrets are configured, run the application:

dotnet build
dotnet run

Expected Output

The console will display:

1. Agent setup messages for each of the three agents
2. Workflow creation confirmation
3. OpenTelemetry tracing output showing agent execution
4. The final article with research, writing, and review feedback
5. A prompt asking if you want to delete the persistent agent from Microsoft Foundry

Sample Execution Flow

=== Microsoft Agent Framework - Multi-Model Orchestration Demo ===
This demo showcases 3 agents working together:
  1. Researcher (Microsoft Foundry Agent) - Researches topics
  2. Writer (Azure OpenAI) - Writes content based on research
  3. Reviewer (Ollama - llama3.2) - Reviews and provides feedback

Setting up Agent 1: Researcher (Microsoft Foundry Agent)...
Setting up Agent 2: Writer (Azure OpenAI)...
Setting up Agent 3: Reviewer (Ollama)...
Creating workflow: Researcher -> Writer -> Reviewer

Starting workflow with topic: 'artificial intelligence in healthcare'
================================================================================

[OpenTelemetry traces shown here...]

=== Final Output ===
[Reviewed article with feedback...]

================================================================================
Workflow completed successfully!

=== Clean Up ===
Do you want to delete the Researcher agent in Microsoft Foundry? (yes/no)

Project Structure

MAF-MultiAgents/
├── Program.cs                      # Main workflow orchestration
├── AIFoundryAgentsProvider.cs      # Microsoft Foundry agent factory
├── ChatClientProvider.cs           # Multi-provider chat client factory
├── AppConfigurationService.cs      # Configuration management
├── MAF-MultiAgents.csproj      # Project dependencies
└── README.md                       # This file

Key Components

AIFoundryAgentsProvider

Creates and manages persistent agents in Microsoft Foundry:

• CreateAIAgent() - Creates a new persistent agent
• DeleteAIAgentInAIFoundry() - Removes agents from Microsoft Foundry

ChatClientProvider

Factory class for creating chat clients with different providers:

• GetChatClient() - Returns a chat client based on available configuration (Azure OpenAI)
• GetChatClientOllama() - Returns an Ollama chat client for local inference

AppConfigurationService

Centralized configuration service that reads from:

• User secrets
• Environment variables

Customization

Change the Topic

Edit the topic variable in Program.cs:

var topic = "artificial intelligence in healthcare";

Modify Agent Instructions

Update the Instructions property for any agent:

AIAgent researcher = new ChatClientAgent(
    azureChatClient,
    new ChatClientAgentOptions
    {
        Name = "Researcher",
        Instructions = "Your custom instructions here..."
    });

Use Different Models

Change the model for any provider:

• Azure OpenAI: Update deploymentName secret with your deployment name
• Ollama: Modify the model parameter in GetChatClientOllama() call

Add More Agents

Extend the workflow by adding additional agents:

AIAgent seoOptimizer = new ChatClientAgent(
    chatClient,
    new ChatClientAgentOptions
    {
        Name = "SEO Optimizer",
        Instructions = "Optimize the article for search engines."
    });

Workflow workflow = AgentWorkflowBuilder
    .BuildSequential(researcher, writer, reviewer, seoOptimizer);

Troubleshooting

Common Issues

Issue: "Azure CLI credentials not found"

• Solution: Run az login to authenticate with Azure CLI

Issue: "Ollama connection refused"

• Solution: Ensure Ollama is running (ollama run llama3.2) and accessible at http://localhost:11434/

Issue: "Model deployment not found"

• Solution: Verify your deploymentName secret matches an actual deployment in your Azure OpenAI or AI Foundry resource

Issue: "GitHub token authentication failed"

• Solution: Verify your GitHub token is valid and has the required permissions

Issue: "Microsoft Foundry endpoint not found"

• Solution: Check that your AZURE_FOUNDRY_PROJECT_ENDPOINT is correct and accessible

Learn More

• Microsoft Agent Framework Documentation
• Microsoft.Extensions.AI Documentation
• Microsoft Foundry
• Ollama Documentation
• OpenTelemetry for .NET

Next Steps

After exploring this sample:

1. Try modifying the agent instructions to see how it affects the output
2. Experiment with different AI models for each agent
3. Add additional agents to create more complex workflows
4. Explore the OpenTelemetry traces to understand agent execution patterns
5. Integrate with other MCP servers for additional capabilities

---

Note: This sample uses preview versions of Microsoft Agent Framework packages. Check for updates regularly as the framework evolves.