Azure/azure-sdk-for-net Azure.AI.AgentServer.Activity_1.0.0-beta.1

on GitHub

1.0.0-beta.1 (2026-11-04)

Features Added

• Initial release of Azure.AI.AgentServer.Activity — a .NET library for hosting a Microsoft 365
  Agents SDK AgentApplication as an Azure AI Foundry hosted agent that speaks the activity
  protocol (POST /activity/messages).
• ActivityServer.Run(...) one-liner startup that wires Kestrel, health probes, OpenTelemetry, and
  the activity endpoint, with overloads for the way your agent is constructed:
  • Run<TAgent>() — host an AgentApplication subclass by type.
  • Run(app => ...) — register handlers inline on the built AgentApplication (no agent class).
  • Run(agentApp) — host a pre-built AgentApplication instance.
  • Run(Func<IServiceProvider, AgentApplication>) — factory registration with access to DI.
  • Run(RequestDelegate) — own the raw request pipeline (the Microsoft 365 Agents SDK is not
    initialized).
• Composition on the shared Core host builder via AgentHostBuilder.AddActivity<TAgent>(),
  AddActivity(agentApp), and AddActivity(Func<IServiceProvider, AgentApplication>).
• Self-hosting on your own WebApplication via AddFoundryActivity() / MapFoundryActivity()
  (aliased as AddActivityServer() / MapActivityServer()), including a RequestDelegate overload
  of MapFoundryActivity for owning the request pipeline in a self-hosted app. This is also the
  two-line conversion path for an existing Microsoft 365 Agents SDK application.
• ActivityServerOptions for configuring the built stack: outbound-auth model (DigitalWorker),
  turn-state Storage, the outbound Connections provider, the Connections:*
ConnectionConfiguration mapping, and a ConfigureServices hook.
• Foundry outbound-auth is configuration-driven: ActivityEnvironment derives the M365
  Connections:* settings from the Foundry-native identity (managed identity for the simple
  agent-instance model, IdentityProxyManager for the digital worker blueprint + FMI exchange
  model) and the SDK-native connection provider mints the outbound reply token. No process
  environment is mutated.
• Session resolution (agent_session_id query / x-agent-session-id header / environment /
  generated), sanitized session-id response header, x-platform-error-source error classification
  (user / platform / upstream), and distributed tracing via ActivityProtocolActivitySource.