Microsoft Agents Hosting Core

The core hosting library for Microsoft 365 Agents SDK. This library provides the fundamental building blocks for creating conversational AI agents, including activity processing, state management, authentication, and channel communication.

This is the heart of the Microsoft 365 Agents SDK - think of it as the engine that powers your conversational agents. It handles the complex orchestration of conversations, manages state across turns, and provides the infrastructure needed to build production-ready agents that work across Microsoft 365 platforms.

What is this?

This library is part of the Microsoft 365 Agents SDK for Python - a comprehensive framework for building enterprise-grade conversational AI agents. The SDK enables developers to create intelligent agents that work across multiple platforms including Microsoft Teams, M365 Copilot, Copilot Studio, and web chat, with support for third-party integrations like Slack, Facebook Messenger, and Twilio.

Release Notes

Packages Overview

We offer the following PyPI packages to create conversational experiences based on Agents:

Package Name	PyPI Version	Description
microsoft-agents-activity		Types and validators implementing the Activity protocol spec.
microsoft-agents-hosting-core		Core library for Microsoft Agents hosting.
microsoft-agents-hosting-aiohttp		Configures aiohttp to run the Agent.
microsoft-agents-hosting-fastapi		Configures fastapi to run the Agent.
microsoft-agents-hosting-msteams		Provides classes to host an Agent for Teams.
microsoft-agents-hosting-dialogs		Dialog system with waterfall dialogs, prompts, and multi-turn conversation management.
microsoft-agents-hosting-slack		Provides classes to host an Agent for Slack.
microsoft-agents-storage-blob		Extension to use Azure Blob as storage.
microsoft-agents-storage-cosmos		Extension to use CosmosDB as storage.
microsoft-agents-authentication-msal		MSAL-based authentication for Microsoft Agents.
microsoft-agents-authentication-entra-auth-sidecar		Credential-free Entra ID Agent ID authentication via the sidecar.

Additionally we provide a Copilot Studio Client, to interact with Agents created in CopilotStudio:

Package Name	PyPI Version	Description
microsoft-agents-copilotstudio-client		Direct to Engine client to interact with Agents created in CopilotStudio

Installation

pip install microsoft-agents-hosting-core

Simple Echo Agent

See the Quickstart sample for full working code.

agents_sdk_config = load_configuration_from_env(environ)

STORAGE = MemoryStorage()
CONNECTION_MANAGER = MsalConnectionManager(**agents_sdk_config)
ADAPTER = CloudAdapter(connection_manager=CONNECTION_MANAGER)
AUTHORIZATION = Authorization(STORAGE, CONNECTION_MANAGER, **agents_sdk_config)

AGENT_APP = AgentApplication[TurnState](
    storage=STORAGE, adapter=ADAPTER, authorization=AUTHORIZATION, **agents_sdk_config
)

@AGENT_APP.activity("message")
async def on_message(context: TurnContext, state: TurnState):
    await context.send_activity(f"You said: {context.activity.text}")

...

start_server(
    agent_application=AGENT_APP,
    auth_configuration=CONNECTION_MANAGER.get_default_connection_configuration(),
)

Core Concepts

AgentApplication vs ActivityHandler

AgentApplication - Modern, fluent API for building agents:

• Decorator-based routing (@agent_app.activity("message"))
• Built-in state management and middleware
• AI-ready with authorization support
• Type-safe with generics

ActivityHandler - Traditional inheritance-based approach:

• Override methods for different activity types
• More familiar to Bot Framework developers
• Lower-level control over activity processing

Route-based Message Handling

@AGENT_APP.message(re.compile(r"^hello$"))
async def on_hello(context: TurnContext, _state: TurnState):
    await context.send_activity("Hello!")


@AGENT_APP.activity("message")
async def on_message(context: TurnContext, _state: TurnState):
    await context.send_activity(f"you said: {context.activity.text}")

Error Handling

@AGENT_APP.error
async def on_error(context: TurnContext, error: Exception):
    # NOTE: In production environment, you should consider logging this to Azure
    #       application insights.
    print(f"\n [on_turn_error] unhandled error: {error}", file=sys.stderr)
    traceback.print_exc()

    # Send a message to the user
    await context.send_activity("The bot encountered an error or bug.")

Key Classes Reference

Core Classes

• AgentApplication - Main application class with fluent API
• ActivityHandler - Base class for inheritance-based agents
• TurnContext - Context for each conversation turn
• TurnState - State management across conversation turns

State Management

• ConversationState - Conversation-scoped state
• UserState - User-scoped state across conversations
• TempState - Temporary state for current turn
• MemoryStorage - In-memory storage (development)

Messaging

• MessageFactory - Create different types of messages
• CardFactory - Create rich card attachments
• InputFile - Handle file attachments

Authorization

• Authorization - Authentication and authorization manager
• ClaimsIdentity - User identity and claims

Quick Links

• 📦 All SDK Packages on PyPI
• 📖 Complete Documentation
• 💡 Python Samples Repository
• 🐛 Report Issues

Sample Applications