3 AI Agents, Real Tools and Models -- All on Azure Container Apps with Foundry

A multi-agent content factory that researches Microsoft technology topics, creates multi-format content, and generates podcasts. Built to familiarize with running and hosting AI Agents on Azure Container Apps, explore agent observability, and register and evaluate agents in Microsoft Foundry.

Lab participants: follow the instructor or check the full lab manual.

Architecture

[Dev UI] --> [Agent 1 - Researcher] --> [Agent 2 - Content Creator] --> [Agent 3 - Podcaster]

Enter a topic (e.g. "Write a comprehensive blog post about Azure Container Apps for developers"). Three agents collaborate:

1. Agent 1 — Tech Research (LangGraph / Python) — Searches Microsoft Learn, Azure Blog, Tech Community, Azure Updates, and GitHub. Uses AI for intent detection, ranks sources by relevance, fetches full content from top hits, follows depth-1 links from trusted domains, and synthesizes a research brief.
2. Agent 2 — Content Creator (Microsoft Agent Framework / .NET) — Transforms the research brief into an original blog post, LinkedIn post, and Twitter thread, all grounded in real sources.
3. Agent 3 — Podcaster (GitHub Copilot SDK / Python) — Creates an engaging podcast script and generates audio. Can use Azure OpenAI TTS or a custom XTTS-v2 server on serverless GPUs.

Dev UI — A lightweight HTML frontend for submitting topics and viewing results from all three agents.

Agents communicate via the A2A (Agent-to-Agent) protocol — each exposes a /.well-known/agent.json card for discovery and a /a2a JSON-RPC endpoint for task submission. Each agent runs as a separate container on Azure Container Apps.

Deploy to Azure

azd up

This creates: Azure AI Foundry (GPT-4o + TTS), ACR, ACA Environment, Log Analytics, Storage Account, and Container Apps for each agent plus the Dev UI.

Run Locally

With Docker Compose:

cd Lab
cp .env.example .env  # fill in your Azure OpenAI credentials
docker compose up     # starts agents 1-3 + dev-ui

Without Docker, see the step-by-step guide: Lab/docs/how-to-run-locally.md.

Verify

curl http://localhost:8001/health   # {"status":"healthy","agent":"research-agent"}
curl http://localhost:8002/health   # {"status":"healthy","agent":"creator-agent"}
curl http://localhost:8003/health   # {"status":"healthy","agent":"podcaster-agent"}

Project Structure

azure.yaml                  # Azure Developer CLI (azd) configuration
Lab/
  docker-compose.yml         # Local multi-container setup
  docs/                      # Architecture docs & run-locally guide
  sample-output/             # Example blog and podcast transcript
  src/
    agent-research/          # Agent 1: Python, LangGraph, FastAPI
    agent-creator/           # Agent 2: .NET 10, Microsoft Agent Framework
    agent-podcaster/         # Agent 3: Python, GitHub Copilot SDK, TTS
    tts-server/              # GPU XTTS-v2 server (full mode only)
    dev-ui/                  # Static HTML + nginx (port 8080)
infra/
  main.bicep                 # Azure deployment (Foundry, ACR, ACA, Storage, GPU)
  main.parameters.json       # azd parameter wiring
  pre-rendered/              # Standalone Bicep for lab vendor provisioning

Tech Stack

Component	Technology
Agent 1 — Research	Python 3.11+, LangGraph, LangChain, FastAPI, httpx, BeautifulSoup
Agent 2 — Creator	.NET 10, Microsoft Agent Framework, ASP.NET Minimal APIs
Agent 3 — Podcaster	Python 3.10+, GitHub Copilot SDK, pydub, httpx
TTS (lab mode)	Azure OpenAI tts-1
TTS (full mode)	Coqui XTTS-v2 on GPU, Azure OpenAI fallback
Protocol	A2A (Agent-to-Agent)
Infrastructure	Azure Container Apps, Azure Container Registry, Bicep
Observability	OpenTelemetry, Azure Application Insights
LLM	Azure OpenAI GPT-4o

Questions?

Create an issue in the repo with your question or concern.

Contributing

This project welcomes contributions and suggestions. Most contributions require you to agree to a Contributor License Agreement (CLA) declaring that you have the right to, and actually do, grant us the rights to use your contribution. For details, visit Contributor License Agreements.

When you submit a pull request, a CLA bot will automatically determine whether you need to provide a CLA and decorate the PR appropriately (e.g., status check, comment). Simply follow the instructions provided by the bot. You will only need to do this once across all repos using our CLA.

This project has adopted the Microsoft Open Source Code of Conduct. For more information see the Code of Conduct FAQ or contact opencode@microsoft.com with any additional questions or comments.

Trademarks

This project may contain trademarks or logos for projects, products, or services. Authorized use of Microsoft trademarks or logos is subject to and must follow Microsoft's Trademark & Brand Guidelines. Use of Microsoft trademarks or logos in modified versions of this project must not cause confusion or imply Microsoft sponsorship. Any use of third-party trademarks or logos are subject to those third-party's policies.