Apps on Azure Blog

5 MIN READ

Build Long-Running AI Agents on Azure App Service with Microsoft Agent Framework

jordanselig

Microsoft

Oct 21, 2025

Learn how to build intelligent AI agents using Microsoft Agent Framework and Azure App Service with async patterns for complex, long-running workflows.

The AI landscape is evolving rapidly, and with the introduction of Microsoft Agent Framework, developers now have a powerful platform for building sophisticated AI agents that go far beyond simple chat completions. These agents can execute complex, multi-step workflows with persistent state, conversation threads, and structured execution—capabilities that are essential for production AI applications.

Today, we're excited to share how Azure App Service provides an excellent platform for running Agent Framework workloads, especially those involving long-running operations. Let's explore why App Service is a great choice and walk through a practical example.

🔗 Quick link to sample app GitHub repo: https://github.com/Azure-Samples/app-service-agent-framework-travel-agent-dotnet

The Challenge: Long-Running Agent Framework Flows

Agent Framework enables AI agents to perform complex tasks that can take significant time to complete:

Multi-turn reasoning: Iterative calls to large language models (LLMs) where each response informs the next prompt
Tool integration: Function calling and external API interactions for real-time data
Complex processing: Budget calculations, content optimization, multi-phase generation
Persistent context: Maintaining conversation state across multiple interactions

These workflows often take 30 seconds to several minutes to complete—far too long for synchronous HTTP request handling. Traditional web applications run into several constraints:

⏱️ Timeout Limitations: HTTP requests have timeout constraints (typically 30-230 seconds)

⚠️ Connection Issues: Clients may disconnect due to network interruptions or browser navigation

📈 Scalability Concerns: Long-running requests block worker threads and don't survive app restarts

🎯 Poor User Experience: Users see endless loading spinners with no progress feedback

The Solution: Async Pattern with App Service

Azure App Service provides a robust solution through the asynchronous request-reply pattern combined with background processing:

API immediately returns (202 Accepted) with a task ID
Background worker processes the Agent Framework workflow
Client polls for status with real-time progress updates
Durable state storage (Cosmos DB) maintains task status and results

This pattern ensures:

✅ No HTTP timeouts—API responds in milliseconds

✅ Resilient to restarts—state survives deployments and scale events

✅ Progress tracking—users see real-time updates (10%, 45%, 100%)

✅ Better scalability—background workers process independently

Rapid Innovation Support

The AI landscape is changing at an unprecedented pace. New models, frameworks, and capabilities are released constantly. Azure App Service's managed platform ensures your applications can adapt quickly without infrastructure rewrites:

Framework Updates: Deploy new Agent Framework SDK versions like any application update
Model Upgrades: Switch between GPT-4, GPT-4o, or future models with configuration changes
Scaling Patterns: Start with combined API+worker, split into separate apps as needs grow
New Capabilities: Integrate emerging AI services without changing hosting infrastructure

App Service handles the platform complexity so you can focus on building great AI experiences.

Sample Application: AI Travel Planner

To demonstrate this pattern, we've built a Travel Planner application that uses Agent Framework to generate detailed, multi-day travel itineraries. The agent performs complex reasoning including:

Researching destination attractions and activities
Optimizing daily schedules based on location proximity
Calculating detailed budget breakdowns
Generating personalized travel tips and recommendations

The entire application runs on a single P0v4 App Service with both the API and background worker combined—showcasing App Service's flexibility for hosting diverse workload patterns in one deployment.

Key Architecture Components

Azure App Service (P0v4 Premium)

Hosts both REST API and background worker in a single app
"Always On" feature keeps background worker running continuously
Managed identity for secure, credential-less authentication

Azure Service Bus

Decouples API from long-running Agent Framework processing
Reliable message delivery with automatic retries
Dead letter queue for error handling

Azure Cosmos DB

Stores task status with real-time progress updates
Automatic 24-hour TTL for cleanup
Rich query capabilities for complex itinerary data

Azure AI Foundry

Hosts persistent agents with conversation threads
Structured execution with Agent Framework runtime
GPT-4o model for intelligent travel planning

One of the powerful features of using Azure AI Foundry with Agent Framework is the ability to inspect agents and conversation threads directly in the Azure portal. This provides valuable visibility into what's happening during execution.

Viewing Agents and Threads in Azure AI Foundry

When you submit a travel plan request, the application creates an agent in Azure AI Foundry. You can navigate to your AI Foundry project in the Azure portal to see:

Agents

The application creates an agent for each request
Important: Agents are **automatically deleted** after the itinerary is generated to keep your project clean
Tip: You'll need to be quick! Navigate to Azure AI Foundry right after submitting a request to see the agent in action
Once processing completes, the agent is removed as part of the cleanup process

Conversation Threads

Unlike agents, threads persist even after the agent completes
You can view the complete conversation history at any time
See the exact prompts sent to the model and the responses generated
Useful for debugging, understanding agent behavior, and improving prompts

The ephemeral nature of agents (created per request, deleted after completion) keeps your Azure AI Foundry project clean while the persistent threads give you full traceability of every interaction.

Get Started Today

The complete Travel Planner application is available as a reference implementation so you can quickly get started building your own apps with Agent Framework on App Service.

🔗 GitHub Repository: https://github.com/Azure-Samples/app-service-agent-framework-travel-agent-dotnet

The repo includes:

Complete .NET 9 source code with Agent Framework integration
Infrastructure as Code (Bicep) for automated deployment
Web UI with real-time progress tracking
Comprehensive README with deployment instructions

Deploy in minutes:

git clone https://github.com/Azure-Samples/app-service-agent-framework-travel-agent-dotnet.git
cd app-service-agent-framework-travel-agent-dotnet
azd auth login
azd up

Key Takeaways

✅ Agent Framework enables sophisticated AI agents beyond simple chat completions

✅ Long-running workflows (30s-minutes) require async patterns to avoid timeouts

✅ App Service provides a simple, cost-effective platform for these workloads

✅ Async request-reply pattern with Service Bus + Cosmos DB ensures reliability

✅ Rapid innovation in AI is supported by App Service's adaptable platform

Whether you're building travel planners, document processors, research assistants, or other AI-powered applications, Azure App Service gives you the flexibility and reliability you need—without the complexity of container orchestration or function programming models.

What's Next? Build on This Foundation

This Travel Planner is just the starting point—a foundation to help you understand the patterns and architecture. Agent Framework is designed to grow with your needs, making it easy to add sophisticated capabilities with minimal effort:

🛠️ Add Tool Calling

Connect your agent to real-time APIs for weather, flight prices, hotel availability, and actual booking systems. Agent Framework's built-in tool calling makes this straightforward.

🤝 Implement Multi-Agent Systems

Create specialized agents (flight expert, hotel specialist, activity planner) that collaborate to build comprehensive travel plans. Agent Framework handles the orchestration.

🧠 Enhance with RAG

Add retrieval-augmented generation to give your agent deep knowledge of destinations, local customs, and insider tips from your own content library.

📊 Expand Functionality

Real-time pricing and availability
Interactive refinement based on user feedback
Personalized recommendations from past trips
Multi-language support for global users

The beauty of Agent Framework is that these advanced features integrate seamlessly into the pattern we've built. Start with this sample, explore the Agent Framework documentation, and unlock powerful AI capabilities for your applications!