Agent Framework Aspire

This repository contains a sample implementation of the Microsoft Agent Framework using Aspire, demonstrating how to build Retrieval-Augmented Generation (RAG) applications with both .NET and Python agents.

Features

• Building agent with .NET and Agent Framework
• Building agent with Python and Agent Framework
• Agent orchestration with .NET and Agent Framework
• Inter-agent communication (A2A) with .NET and Agent Framework
• Using MCP with .NET and Agent Framework
• Using function filtering with Agent Framework
• Test agents behaviour without using evaluation frameworks
• Using Aspire to host multi-agent applications

Run the sample

This sample requires .Net 10 Preview SDK and Python 3.11+ installed on your machine.

To allow Aspire to create or reference existing resources on Azure (e.g. Foundry), you need to configure Azure settings in the appsettings.json file:

"Azure": {
  "SubscriptionId": "<YOUR-SUBSCRIPTION-ID>",
  "AllowResourceGroupCreation": false,
  "Location": "<YOUR-LOCATION>",
  "CredentialSource": "AzureCli"
}

If you want to run the Python agent alongside the .NET agent, you need to install the dependencies using uv:

cd src/agents-python
uv sync --prerelease=allow

Use aspire cli to run the sample.

Powershell:

Invoke-RestMethod https://aspire.dev/install.ps1 -OutFile aspire-install.ps1
./aspire-install.ps1 -q dev

aspire run

Bash:

curl -sSL https://aspire.dev/install.sh -o aspire-install.sh
./aspire-install.sh -q dev

aspire run

To ease the debug experience, you can use the Aspire extension for Visual Studio Code. Otherwise, you can use the C# Dev Kit for Visual Studio Code.

Aspire single-file AppHost

This sample can be use with single-file AppHost. Change the aspire configuration in the .aspire/settings.json file to point to the apphost.cs file:

{
  "features": {
    "singlefileAppHostEnabled": "true",
    "minimumSdkCheckEnabled": "false"
  },
  "appHostPath": "../src/apphost-singlefile/apphost.cs"
}

Then run the sample as usual with aspire run.

Folder Structure

src/
├── frontend/                   # React Frontend application
├── agents-python/              # Python API for RAG
│   ├── services/               # Services for Python Agent
│   └── tools/                  # Tools for Python Agent
├── agents-dotnet/              # dotnet Agent API
│   ├── services/               # Services for dotnet Agent
│   └── tools/                  # Tools for dotnet Agent
├── groupchat-dotnet/           # dotnet Group Chat API
├── mcp-server-dotnet/          # dotnet MCP Server
├── custom-workflow-python/     # Custom Python Workflow
├── apphost/                    # Aspire App Host
├── apphost-singlefile/         # Aspire App Host (Single File)
└── service-defaults/           # Default configurations for services
test/
└── agents-dotnet-tests/        # Test project for dotnet agent

.NET Agent

Is a .NET-based agent hosted in a minimal API.

Role: Document management, policy lookup, and compliance Local Tools (methods returning mocked data):

• PDF text extraction and search
• Word/PowerPoint document parsing
• SharePoint/file system document indexing
• Policy and procedure lookups
• Compliance rule checking
• Document version management

Data Sources:

• Company policies, HR handbook, compliance documents
• Project documentation, technical specifications
• Legal contracts, vendor agreements
• Training materials, standard operating procedures

Sample Questions:

This question will invoke the remote mcp tool":

• "What's our remote work policy?"

While these questions will invoke local tools:

• "Find the latest safety procedures for warehouse operations"
• "What are the approval requirements for purchases over $5000?"

Python Agent

Is a Python-based agent hosted in a uv-based uvicorn application.

Role: Financial analysis, reporting, and business metrics Local Tools (methods returning mocked data):

• CSV/Excel file parsing and analysis
• Database queries (SQLite, PostgreSQL local connections)
• Financial calculations and trend analysis
• Report generation (PDF, charts)
• Data validation and cleaning
• Statistical analysis libraries (pandas, numpy)

Data Sources:

• Sales reports, revenue data, expense tracking
• Employee performance metrics, payroll data
• Inventory levels, supply chain data
• Customer demographics and behavior analytics

Sample Questions:

• "What were our top-performing products last quarter?"
• "Show me the sales trend for the past 6 months"
• "Calculate our customer acquisition cost"

Dotnet Group Chat

Is a .NET-based group chat that creates a local agent with the same capabilities of the Python Agent and reference the .NET agent via A2A. It can be easily invoked with a hardcoded prompt for convenience from the Aspire dashboard. It can also be invoked via the frontend.

Sample Questions:

• According to our procurement policy, what vendors are we required to use for office supplies, and what has been our spending pattern with those vendors over the past 6 months?

Python Custom Workflow

This is a custom workflow written in python that implements a custom flow using both dotnet and python agent via A2A. It can be easily invoked with a hardcoded prompt for convenience from the Aspire dashboard. The workflow is described here

Test project

The assumption in the test project is that the agents will respond correctly if they invoke the correct tools with the correct parameters. I use function filters to gather the tools invocations and parameters and validate them against the expected ones. This way I can test the agents behaviour without using evaluation frameworks.