Azure Container Apps Multi-Agent Workflow: Building an Observable Content Factory

Microsoft's open-source Azure Container Apps Multi-Agent Content Factory project demonstrates how to run and host AI agents on Azure Container Apps, enabling a complete workflow for research, content creation, and podcast generation. This project is a production-ready architecture template that covers key aspects such as security and observability, helping developers familiarize themselves with best practices for multi-agent systems.

With the improvement of large model capabilities, multi-agent collaboration has become an important direction in AI application architecture. Aligning with this trend, this project provides a complete reference implementation of a multi-agent content factory. It is positioned to help developers familiarize themselves with best practices for running AI agents on Azure Container Apps, explore agent observability solutions, and conduct registration evaluations in Microsoft Foundry. It is a production-ready architecture template.

The content factory consists of four core components:

1. Orchestrator and UI: Coordinates agent workflows, receives user configurations and prompts;
2. Research Agent: Built on LangGraph/Python, responsible for topic research and generating briefings, using ACA sandbox to ensure security;
3. Creation Agent: Built on Microsoft Agent Framework/.NET, converts research briefings into original content;
4. Podcast Agent: Built on GitHub Copilot SDK/Python, creates podcast scripts and generates audio (Azure OpenAI TTS).

Technical highlights include:

• Multi-framework coexistence: Agents are built using LangGraph, Microsoft Agent Framework, and GitHub Copilot SDK;
• Security: AI code is executed in ACA sandbox, outbound traffic is restricted to allowlists;
• Observability: End-to-end monitoring and tracing via Application Insights;
• Model inference: Connects to Microsoft Foundry, supporting models like GPT-4o and TTS.

The project provides two scenarios:

1. Football match prediction: Research agent collects data, creation agent generates prediction articles, podcast agent generates preview audio;
2. Microsoft technology research: Research agent mines technical documents, creation agent generates interpretation articles, podcast agent generates technical podcasts.

Supports local running (for development and debugging) and Azure deployment (AZD UP one-click resource configuration: Container Apps, Registry, Application Insights, Foundry connection). After deployment, enter a topic via the web interface to observe the agent collaboration process.

The project's value is reflected in:

1. Secure operation: Demonstrates enterprise-level AI agent security measures;
2. Observability: Provides operation and maintenance monitoring solutions for agents in production environments;
3. Multi-framework collaboration: Demonstrates heterogeneous agent collaboration;
4. Practical tool: Can build automated content creation pipelines, suitable for scenarios like media and marketing.

This project is a production-ready reference implementation of a multi-agent system, covering key aspects such as security and observability. As AI agent technology evolves, such reference implementations will help enterprises move from proof of concept to production. Microsoft contributes engineering practice experience through open source to promote the implementation of multi-agent systems.