TL;DR
Microsoft published the Agent Framework, an open-source SDK for building LLM-powered agents and graph-based multi-agent workflows for .NET and Python. The project merges ideas from Semantic Kernel and AutoGen, is available in public preview, and includes migration guides and GitHub feedback channels.
What happened
Microsoft released the Agent Framework, an open-source development kit intended as a successor to Semantic Kernel and AutoGen. The framework targets .NET and Python developers and provides two main capability areas: AI agents (single agents that use LLMs to interpret input, call tools or MCP servers, and produce responses) and graph-based workflows (which coordinate multiple agents and functions for multi-step or long-running processes). The kit supplies core building blocks such as model clients, an agent thread for state management, context providers for memory, middleware hooks, and MCP clients for tool integration. Workflows support type-based routing, nesting, checkpointing, request/response patterns for human-in-the-loop use, and several multi-agent orchestration patterns. Microsoft describes the project as being in public preview, offers installation packages for Python and .NET, and asks the community to file feedback and contributions via its GitHub repository. The documentation also warns developers to manage data shared with third-party servers carefully.
Why it matters
- Unifies features from Semantic Kernel and AutoGen into a single, extensible SDK for agent-based applications.
- Provides built-in workflow primitives and state management aimed at long-running and human-in-the-loop scenarios.
- Supports multiple model providers and runtime environments, lowering integration friction for existing teams.
- Open-source status and GitHub feedback channels invite community contributions and faster iteration.
Key facts
- Agent Framework is open-source and intended for .NET and Python developers.
- It combines concepts from Semantic Kernel and AutoGen and is positioned as their successor.
- Two primary feature areas: AI agents (LLM-driven single agents) and graph-based workflows (multi-agent orchestration).
- Core components include model clients, an agent thread for state, context providers, middleware, and MCP clients for tool access.
- Workflows support type-based routing, nesting, checkpointing, request/response patterns, and several orchestration patterns.
- Supported model providers include Azure OpenAI, OpenAI, and Azure AI (per the documentation).
- Installation examples: pip install agent-framework –pre for Python; dotnet add package Microsoft.Agents.AI for .NET.
- The project is listed as public preview and the team asks for feedback or issues to be filed on GitHub.
- Documentation includes migration guides for projects using Semantic Kernel or AutoGen.
What to watch next
- Progress toward general availability and a published roadmap — not confirmed in the source
- Community activity and contributions on the project's GitHub repository (issues, pull requests and feedback) — confirmed in the source
- Updates to migration guides and examples to simplify porting from Semantic Kernel or AutoGen — not confirmed in the source
- Guidance or tooling from Microsoft addressing safe data handling and third‑party server integrations — not confirmed in the source
Quick glossary
- AI agent: A software actor that uses a large language model to interpret inputs, call tools or services, and generate responses or take actions.
- Workflow: A graph-based sequence of operations that coordinates agents and functions to perform multi-step or long-running tasks.
- LLM (Large Language Model): A type of machine learning model trained on large text corpora that can generate or transform language-based outputs.
- Middleware: Code that intercepts and can modify agent actions or messages as they flow through the system.
- Checkpointing: Saving the state of a workflow so execution can be resumed or recovered later.
Reader FAQ
Is Microsoft Agent Framework open-source?
Yes — the documentation describes it as an open-source development kit and invites contributions on GitHub.
Which programming languages are supported?
The framework targets .NET and Python.
Which model providers does it support?
Documentation lists Azure OpenAI, OpenAI, and Azure AI as supported model providers.
Is the framework production-ready?
The project is currently in public preview, per the documentation.
How do I install the SDK?
Python: pip install agent-framework –pre. .NET: dotnet add package Microsoft.Agents.AI.
Free to join. Request to attend. Microsoft AI Tour Take your business to the AI frontier. Learn Microsoft Agent Framework Ask Learn Microsoft Agent Framework In this article Why another…
Sources
- Microsoft Agent Framework
- Introducing Microsoft Agent Framework | Microsoft Azure Blog
- Introducing Microsoft Agent Framework: The Open-Source …
- Microsoft Announces Open-Source Agent Framework to …
Related posts
- Microsoft Patch Tuesday December 2025: 56 Fixes Including One Zero‑Day
- Microsoft study finds Copilot chats embedding into daily user routines
- Microsoft December 2025 security update disrupts MSMQ on older Windows systems