The Durable Task extension for Microsoft Agent Framework brings durable execution directly into the Microsoft Agent Framework. You can register agents with the extension to make them automatically durable — with persistent sessions, built-in API endpoints, and distributed scaling — without changes to your agent logic.
The extension internally implements entity-based agent loops, where each agent session is a durable entity that automatically manages conversation state and checkpointing.
| Capability | Description |
|---|---|
| Persistent sessions | Conversation history survives restarts, crashes, and scaling events — no external database needed |
| Automatic checkpointing | Every agent interaction and tool call is checkpointed; completed work is never re-executed on recovery |
| Built-in HTTP endpoints | Send messages, check status, and manage sessions via auto-generated REST APIs |
| Multi-agent orchestration | Coordinate multiple specialized agents as steps in a durable orchestration with automatic recovery |
| Graph-based workflows | Define multi-step pipelines of executors and agents using WorkflowBuilder with automatic checkpointing |
| Session TTL | Automatic cleanup of idle sessions to manage storage and costs |
| Two hosting options | Azure Functions (serverless) or bring-your-own compute (console apps, containers, etc.) |
One line to make your agent durable with serverless hosting:
using IHost app = FunctionsApplication
.CreateBuilder(args)
.ConfigureFunctionsWebApplication()
.ConfigureDurableAgents(options => options.AddAIAgent(agent))
.Build();
app.Run();Host the agent with the Durable Task SDK directly — no Azure Functions dependency:
IHost host = Host.CreateDefaultBuilder(args)
.ConfigureServices(services =>
{
services.ConfigureDurableAgents(
options => options.AddAIAgent(agent),
workerBuilder: builder => builder.UseDurableTaskScheduler(connectionString),
clientBuilder: builder => builder.UseDurableTaskScheduler(connectionString));
})
.Build();
await host.StartAsync();These samples show how to host agents with the Durable Task extension. Each agent session is a durable entity that persists conversation history, supports tool calling, and recovers automatically from failures.
| Pattern | Description |
|---|---|
| Single agent | One LLM agent with persistent sessions and built-in HTTP endpoints |
| Multi-agent orchestration | Multiple specialized agents coordinated as checkpointed steps in a durable orchestration |
| Tool calling | Agents with function tools — tool calls are checkpointed and not re-executed on recovery |
| MCP server | Agent hosted as a Model Context Protocol server |
| Reliable streaming | Real-time token streaming with durable delivery guarantees |
| Human-in-the-loop | Agents that pause for human approval before continuing |
These samples show how to use the Microsoft Agent Framework WorkflowBuilder with the Durable Task extension. The extension automatically checkpoints each step in the graph and recovers from failures without changes to the workflow definition.
| Pattern | Description |
|---|---|
| Sequential | Chain executors into a multi-step pipeline (e.g., look up → cancel → notify) |
| Fan-out/fan-in | Run multiple agents or executors in parallel, then aggregate results |
| Conditional routing | Route execution to different branches based on runtime results (e.g., spam detection) |
| Human-in-the-loop | Pause workflow execution at designated points to wait for external approval |
| Sub-workflows | Compose complex workflows from reusable sub-workflows |
| Shared state | Pass state between workflow steps using context |
| Events | React to external events during workflow execution |
python/
hosting/
azure-functions/ → Azure Functions agent hosting samples
durable-task/ → Durable Task SDK agent hosting samples (no Azure Functions)
dotnet/
hosting/
azure-functions/ → Azure Functions agent hosting samples
console-apps/ → Console app agent hosting samples
durable-maf-workflows/
azure-functions/ → Graph-based workflow samples (Azure Functions)
console-apps/ → Graph-based workflow samples (Console Apps)
Note: These directories are symlinks into the
microsoft/agent-frameworkrepo, which is included as a Git submodule atexternal/agent-framework. This avoids duplicating samples across repos.
- Docker — for the Durable Task Scheduler emulator
- .NET 10 SDK (for .NET samples) or Python 3.10+ (for Python samples)
- (Optional) Azure OpenAI endpoint for real LLM integration
Clone this repo with submodules to pull the sample code:
git clone --recurse-submodules https://github.com/Azure-Samples/Durable-Task-Scheduler.gitIf you've already cloned without submodules, initialize them separately:
git submodule update --init external/agent-frameworkThen start the Durable Task Scheduler emulator:
docker run -d -p 8080:8080 -p 8082:8082 mcr.microsoft.com/dts/dts-emulator:latestThe dashboard is available at http://localhost:8082.
To pull the latest samples from the agent-framework repo:
cd external/agent-framework
git pull origin main
cd ../..
git add external/agent-framework
git commit -m "Update agent-framework submodule"A GitHub Actions workflow also runs weekly to auto-update the submodule via PR.
| Sample | Hosting | Description |
|---|---|---|
| Single Agent | Azure Functions | Basic durable agent with persistent sessions |
| Orchestration Chaining | Azure Functions | Multi-agent sequential orchestration |
| Orchestration Concurrency | Azure Functions | Parallel agent execution with fan-out/fan-in |
| Orchestration Conditionals | Azure Functions | Conditional routing between agents |
| Human-in-the-Loop | Azure Functions | Agent pauses for human approval |
| ReliableStreaming | Azure Functions | Real-time token streaming with durability |
| MCP Server | Azure Functions | Agent as a Model Context Protocol server |
| Custom State | Azure Functions | Agent with custom persistent state |
| Single Agent | Console App | Basic durable agent without Azure Functions |
| Orchestration Chaining | Console App | Multi-agent orchestration in a console app |
| Orchestration Concurrency | Console App | Parallel agents in a console app |
| Orchestration Conditionals | Console App | Conditional agent routing in a console app |
| Human-in-the-Loop | Console App | Human approval in a console app |
| Custom State | Console App | Custom agent state in a console app |
| Durable Agent Client | Console App | Client for interacting with durable agents |
| Sample | Hosting | Description |
|---|---|---|
| Sequential | Azure Functions | Order cancellation pipeline: look up → cancel → notify |
| Concurrent | Azure Functions | Fan-out to physicist & chemist agents, fan-in to aggregate |
| Human-in-the-Loop | Azure Functions | Expense reimbursement with manager + parallel finance approvals |
| MCP Tool | Azure Functions | Workflow exposed as an MCP tool |
| Combined | Azure Functions | Workflows + agents in the same app |
| Sequential | Console App | Sequential executor pipeline |
| Concurrent | Console App | Fan-out/fan-in with parallel execution |
| Conditional Edges | Console App | Runtime routing based on conditions |
| Events | Console App | Reacting to external events in workflows |
| Shared State | Console App | Passing state between workflow steps |
| Sub-Workflows | Console App | Composing reusable sub-workflows |
| Human-in-the-Loop | Console App | Workflow pauses for external approval |
| Streaming | Console App | Streaming workflow events |
| Sample | Hosting | Description |
|---|---|---|
| Single Agent | Azure Functions | Basic durable agent with persistent sessions |
| Multi-Agent Orchestration | Azure Functions | Multiple agents coordinated in an orchestration |
| Tool Calling | Azure Functions | Agent with function tools |
| Human-in-the-Loop | Azure Functions | Agent pauses for human approval |
| Single Agent | Durable Task SDK | Agent hosted with the DT SDK directly |
| Multi-Agent Orchestration | Durable Task SDK | Multi-agent orchestration without Azure Functions |
| Tool Calling | Durable Task SDK | Tool-calling agent with the DT SDK |
| Human-in-the-Loop | Durable Task SDK | Human approval with the DT SDK |
- Durable Task extension for Microsoft Agent Framework — Full documentation
- Durable Task for AI Agents — Why durable execution matters for AI agents
- Agentic Application Patterns — All supported patterns
- Microsoft Agent Framework — The agent framework itself
- Durable Task Scheduler Dashboard — Monitor agents, orchestrations, and workflows