介绍 Microsoft Agent Framework：用于智能体AI应用的开源引擎

Why agents need a new foundation    为何智能体需要新基石

Over the last year, developers have been experimenting with AI agents in every imaginable form. Agents are not just chatbots or copilots — they are autonomous software components that can reason about goals, call tools and APIs, collaborate with other agents, and adapt dynamically. Whether it’s a retrieval agent for research, a coding agent embedded in a dev workflow, or a compliance agent ensuring policy enforcement, agents are becoming the next layer of application logic.

在过去的一年里，开发者们以各种可能的形式对AI智能体进行了实验。智能体不仅仅是聊天机器人或助手，它们是能够自主推理目标、调用工具和API、与其他智能体协作并动态适应环境的自主软件组件。无论是用于研究的检索智能体、嵌入开发流程中的编码智能体，还是确保政策执行的合规智能体，智能体渐成为应用逻辑的新层级。

Yet despite the excitement, the path from prototype to production has been fraught with obstacles. Many of the most popular open-source frameworks are fragmented, each with their own APIs and abstractions. Local development rarely maps cleanly to cloud deployments. And most importantly, enterprise readiness is missing: observability, compliance hooks, security, and long-running durability are table stakes in OSS frameworks.

尽管前景令人振奋，但从原型到生产的道路却布满荆棘。许多热门开源框架各自为政，均拥有独立的API与抽象体系。本地开发环境很难与云端部署无缝对接。更重要的是，企业级需求尚未满足：可观测性、合规性接口、安全防护与长期运行稳定性在开源框架中仍处于基础能力缺失状态。

At Microsoft, we’ve had a front-row seat to this problem. With Semantic Kernel, we gave developers a stable SDK with connectors into enterprise systems, content moderation, and telemetry. With AutoGen, pioneered in Microsoft Research, we opened the door to experimental multi-agent orchestration patterns that inspired the community. Both had passionate users — but each had gaps.

在微软，我们一直密切关注这一问题。通过Semantic Kernel，我们为开发者提供了包含企业系统连接器、内容审核与遥测技术的稳定SDK。而源自微软研究院的AutoGen，则开创性地开启了实验性多智能体编排模式的大门，为社区带来启发。这两个项目都拥有忠实用户群体，但各自存在局限性。

Developers asked us: why can’t we have both — the innovation of AutoGen and the trust and stability of Semantic Kernel — in one unified framework?

开发者们向我们发出疑问：为何不能在一个统一框架中，同时拥有AutoGen的创新性与Semantic Kernel的可靠稳定？

That’s exactly why we built the Microsoft Agent Framework.

这正是我们构建Microsoft Agent Framework的原因。

Introducing Microsoft Agent Framework    介绍 Microsoft Agent Framework

Microsoft Agent Framework is an open-source SDK and runtime designed to let developers build, deploy, and manage sophisticated multi-agent systems with ease. It unifies the enterprise-ready foundations of Semantic Kernel with the innovative orchestration of AutoGen, so teams no longer have to choose between experimentation and production.

Microsoft Agent Framework 是一个开源的 SDK 和运行时环境，旨在让开发者能够轻松地构建、部署和管理复杂的多智能体系统。它将 Semantic Kernel 的企业级就绪基础与 AutoGen 的创新性编排能力融为一体，使团队无需再在实验探索与生产应用之间做出取舍。

	Semantic Kernel	AutoGen	Microsoft Agent Framework
Focus 重点	Stable SDK with enterprise connectors, workflows, and observability 具备企业连接器、工作流和可观察性的稳定 SDK	Experimental multi-agent orchestration from research 来自研究领域的实验性多智能体编排	Unified SDK combining innovation + enterprise readiness 结合创新性与企业级就绪的统一 SDK
Interop 互操作性	Plugins, connectors, and support for MCP, A2A, OpenAPI 插件、连接器，支持 MCP、A2A、OpenAPI	Tool integration supported; lacks standardized cross-runtime protocols 支持工具集成；缺乏标准化的跨运行时协议	Built-in connectors, MCP + A2A + OpenAPI 内置连接器，支持 MCP、A2A 和 OpenAPI
Memory 记忆	Multiple vector store connectors and memory store abstraction (e.g. Elasticsearch, MongoDB) 多种向量存储连接器和记忆存储抽象（如 Elasticsearch、MongoDB）	Support for in-memory / buffer history + external vector store memory options (ChromaDB, Mem0, etc) 支持内存/缓冲历史记录 + 外部向量存储选项（ChromaDB、Mem0 等）	Pluggable memory across stores (first-party and third-party), persistent & adaptive memory stored with retrieval, hybrid appraoches 支持多种存储（第一方和第三方）的可插拔内存，具备持久化、自适应记忆存储与检索功能，支持混合方法
Orchestration 编排	Deterministic + dynamic orchestration (Agent Framework, Process Framework) 确定性 + 动态编排（智能体框架、流程框架）	Dynamic LLM orchestration (debate, reflection, facilitator/worker, group chat) 动态 LLM 编排（辩论、反思、引导者/工作者、群组对话）	Deterministic + dynamic orchestration (Agent Orchestration, Workflow Orchestration) 确定性 + 动态编排（智能体编排、工作流编排）
Enterprise readiness 企业级就绪	Telemetry, observability, compliance hooks 遥测、可观察性、合规性接口	Minimal 极少	Observability, approvals, CI/CD, long-running durability, hydration 可观察性、审批机制、CI/CD、长时间运行的持久性、状态恢复（hydration）

With Microsoft Agent Framework, you get:

通过 Microsoft Agent Framework，您将获得：

• Open standards & interoperability — MCP, A2A, and OpenAPI ensure agents are portable and vendor-neutral.    开放标准与互操作性 —— MCP、A2A 和 OpenAPI 确保智能体具有可移植性且不依赖特定供应商。
• Pipeline for research-to-production — bleeding-edge orchestration patterns from Microsoft Research are now ready for enterprise use.    从研究到生产的流水线 —— 来自微软研究院的前沿编排模式现已可用于企业环境。
• Community-driven extensibility — modular by design, with connectors, pluggable memory, and declarative agent definitions.    社区驱动的可扩展性 —— 设计上模块化，支持连接器、可插拔记忆和声明式智能体定义。
• Enterprise readiness — built-in observability, approvals, security, and long-running durability.    企业级就绪 —— 内置可观察性、审批机制、安全性和长时间运行的持久性。

Microsoft Agent Framework doesn’t replace Semantic Kernel and AutoGen — it builds on them. By consolidating their strengths, it gives developers one foundation to move from experimentation to enterprise deployment without compromise. Microsoft Agent Framework supports both Agent Orchestration (LLM-driven, creative reasoning and decision-making) and Workflow Orchestration (business-logic driven, deterministic multi-agent workflows). Together, they allow teams to choose the right approach for the problem: flexible collaboration for open-ended tasks, or structured workflows for repeatable enterprise processes.

Microsoft Agent Framework 并非取代 Semantic Kernel 和 AutoGen，而是建立在它们的基础之上。通过整合两者的优势，它为开发者提供了一个统一的基础，使其能够无缝地从实验阶段过渡到企业级部署，而无需做出任何妥协。Microsoft Agent Framework 同时支持智能体编排（由大语言模型驱动，侧重于创造性推理与决策）和工作流编排（由业务逻辑驱动，侧重于确定性的多智能体工作流）。这两者相结合，使得团队能够根据具体问题选择合适的方法：为开放性任务采用灵活的协作模式，或为可重复的企业流程采用结构化工作流。

Looking ahead, Microsoft Agent Framework further advances integrations across Microsoft’s agent development stack, including the integration with the Microsoft 365 Agents SDK and a shared runtime with Azure AI Foundry Agent Service. The Microsoft 365 Agents SDK is the pro-code toolkit that lets developers build full-stack, multi-channel agents and publish them across Microsoft 365 Copilot, Teams, web, and other surfaces, with deep interoperability into Copilot Studio’s low-code connectors and Microsoft 365 Copilot custom engine agents. By converging this SDK with Microsoft Agent Framework—and aligning it with the shared runtime used in Foundry Agent Service—developers will gain one unified set of abstractions to create, run, scale, and publish agents. This means you can prototype locally, debug with consistent telemetry, and then seamlessly move into scaled hosting with enterprise-grade observability, compliance, and security—all without rewriting your agents—and then publish them into any communication channels of choice where you want to surface your agents.

展望未来，Microsoft Agent Framework 将进一步深化与微软智能体开发技术栈的集成，包括与 Microsoft 365 Agents SDK 的整合，以及与 Azure AI Foundry Agent Service 共享运行时环境。Microsoft 365 Agents SDK 是一个面向专业开发者的工具包，使开发者能够构建全栈式、多通道的智能体，并将其发布到 Microsoft 365 Copilot、Teams、网页以及其他平台。它还能深度对接 Copilot Studio 的低代码连接器和 Microsoft 365 Copilot 自定义引擎智能体。

通过将此 SDK 与 Microsoft Agent Framework 融合，并与 Foundry Agent Service 所采用的共享运行时保持一致，开发者将获得一套统一的抽象层，用于创建、运行、扩展和发布智能体。这意味着您可以在本地进行原型设计，使用一致的遥测数据进行调试，然后无缝过渡到具备企业级可观察性、合规性和安全性的规模化托管环境，而无需重写您的智能体，最终还能将它们发布到您希望呈现智能体的任何通信渠道中。

The Four Pillars of Agent Framework    Agent Framework 的四大支柱

1. Open Standards & Interoperability    开放标准与互操作性

Agents don’t exist in isolation — they need to connect to data, tools, and each other. Microsoft Agent Framework was built with open standards at its core, so developers can choose their integrations and ensure their systems remain portable across frameworks and clouds.

智能体并非孤立存在——它们需要连接数据、工具以及彼此。Microsoft Agent Framework 以开放标准为核心构建，开发者可以自由选择集成方式，并确保其系统在不同框架和云平台之间保持可移植性。

• MCP (Model Context Protocol): Agents can dynamically discover and invoke external tools or data servers exposed over MCP. Microsoft Agent Framework makes it easy to connect to a growing ecosystem of MCP-compliant services without custom glue code.    MCP（模型上下文协议）：智能体可以通过 MCP 动态发现并调用外部工具或数据服务器。Microsoft Agent Framework 让开发者能够轻松连接不断增长的符合 MCP 标准的服务生态，而无需编写自定义的粘合代码。
• Agent-to-Agent (A2A): Agents can collaborate across runtimes using structured, protocol-driven messaging. A2A support allows developers to create workflows where one agent retrieves data, another analyzes it, and a third validates results — even if they’re running in different frameworks or environments.    智能体间通信 (A2A)：智能体可以通过结构化的、基于协议的消息在不同运行时之间协作。A2A 支持使开发者能够创建这样的工作流：一个智能体检索数据，另一个进行分析，第三个验证结果——即使它们运行在不同的框架或环境中。
• OpenAPI-first design: Any REST API with an OpenAPI specification can be imported as a callable tool instantly. Microsoft Agent Framework handles schema parsing, tool definition, and secure invocation so developers can leverage thousands of enterprise APIs without building wrappers by hand.    以 OpenAPI 优先的设计：任何拥有 OpenAPI 规范的 REST API 都能立即导入为可调用工具。Microsoft Agent Framework 负责处理模式解析、工具定义和安全调用，使开发者无需手动构建包装器即可利用成千上万的企业级 API。
• Cloud-agnostic runtime: Agents can run in containers, on-premises, or across multiple clouds, making them portable across environments. Developers can spin up a single agent with their preferred SDK (Azure OpenAI, OpenAI, etc.), add tools by wrapping existing methods as AIFunctions, and immediately connect to external APIs.    云无关的运行时：智能体可以在容器中、本地部署或跨多个云平台运行，从而实现在不同环境间的可移植性。开发者可以使用他们偏好的 SDK（如 Azure OpenAI、OpenAI 等）快速启动单个智能体，通过将现有方法包装为 AIFunctions 来添加工具，并立即连接到外部 API。

The latest update to the VS Code AI Toolkit brings a streamlined experience for building with the Microsoft Agent Framework, enabling developers to locally create, run, and visualize multi-agent workflows. These enhancements simplify the inner dev loop, making it easier to build, debug, and iterate on multi-agent systems within the familiar VS Code environment.

最新的 VS Code AI Toolkit 更新为使用 Microsoft Agent Framework 开发提供了简化的体验，使开发者能够在本地创建、运行和可视化多智能体工作流。这些增强功能简化了内部开发循环，让开发者能够在熟悉的 VS Code 环境中更轻松地构建、调试和迭代多智能体系统。

2. Pipeline for Research    从研究到生产的管道

Microsoft Agent Framework is designed to be the bridge between research innovation and enterprise-ready production. Many of the most exciting breakthroughs in multi-agent orchestration patterns come out of Microsoft Research in AutoGen, and the new framework makes those ideas usable in real-world systems without sacrificing durability, governance, or performance.

Microsoft Agent Framework 旨在成为连接研究创新与企业级生产就绪系统之间的桥梁。许多关于多智能体编排模式的激动人心的突破性进展都源自微软研究院的 AutoGen，而新框架使得这些想法能够在真实系统中应用，同时不牺牲系统的持久性、治理能力或性能。

The framework supports:

该框架支持以下编排模式：

• Sequential orchestration for step-by-step workflows.    顺序编排：用于逐步执行的工作流。
• Concurrent orchestration where agents work in parallel.    并发编排：多个智能体并行工作。
• Group chat orchestration where agents brainstorm collaboratively.    群组对话编排：智能体协同进行头脑风暴。
• Handoff orchestration where responsibility moves between agents as context evolves.    交接编排：随着上下文的变化，任务职责在不同智能体间动态转移。
• Magentic orchestration where a manager agent builds and refines a dynamic task ledger, coordinating specialized agents (and sometimes humans) for complex, open-ended problems.    磁性编排（Magentic orchestration）：由一个管理型智能体构建并完善动态任务账本，协调专业化的智能体（有时也包括人类）来共同解决复杂且开放性的问题。

To serve both innovators and production-minded developers, Microsoft Agent Framework also provides an extension package for experimental features — a clearly labeled incubation channel where advanced users can try out cutting-edge capabilities from Microsoft Research and the open-source community. These features are transparent about their experimental status, while successful innovations graduate naturally into the stable framework.

为同时满足创新探索者与生产导向型开发者的需求，Microsoft Agent Framework 还提供了一个用于实验性功能的扩展包——这是一个明确标注的孵化通道，让高级用户可以试用来自微软研究院和开源社区的前沿功能。这些功能会明确标明其“实验性”状态，而成功的创新成果将自然地融入稳定版框架。

These patterns — once prototypes — now run with durability, auditability, and enterprise controls. It’s the best of research innovation, matured for real-world use.

这些曾经只是原型的模式，如今已能在具备持久性、可审计性和企业级管控的环境下运行。这正是将研究创新的精华，打磨成熟以用于现实世界应用的最佳体现。

3. Extensible by Design & Community-Driven    模块化设计与社区驱动

Microsoft Agent Framework is 100% open source and designed to grow with the community. Its modular design makes it easy to extend, customize, and contribute.

Microsoft Agent Framework 100% 开源，旨在与社区共同成长。其模块化的设计使其易于扩展、定制和贡献。

• Connectors to enterprise systems: Agent Framework inherits a broad set of built-in connectors (Azure AI Foundry, Microsoft Graph, Microsoft Fabric, SharePoint, Oracle, Amazon Bedrock, MongoDB, and a various SaaS system through Azure Logic Apps) so agents can work with enterprise data from day one.    连接企业系统的连接器：Agent Framework 继承了大量内置连接器（如 Azure AI Foundry、Microsoft Graph、Microsoft Fabric、SharePoint、Oracle、Amazon Bedrock、MongoDB，以及通过 Azure Logic Apps 连接的各种 SaaS 系统），使智能体从第一天起就能处理企业数据。
• Pluggable memory modules: Developers can choose Redis, Pinecone, Qdrant, Weaviate, Elasticsearch, Postgres, or their own store for conversational memory. Agent Framework provides the abstraction; you decide the backend.    可插拔记忆模块：开发者可自由选用Redis、Pinecone、Qdrant、Weaviate、Elasticsearch、Postgres或自定义存储方案来管理对话记忆。Agent Framework 提供抽象层，后端存储由您自行决定。
• Declarative agents: YAML or JSON definitions allow developers to specify prompts, roles, and tools declaratively. These files can be version-controlled, templatized, and shared across teams.    声明式智能体：通过 YAML 或 JSON 定义，开发者可以以声明方式指定提示词、角色和工具。这些文件可以进行版本控制、模板化，并在团队间共享。
• Community innovation: Agent Framework is designed to absorb community-driven orchestration strategies, new connectors, and best practices.    社区创新：Agent Framework 旨在吸收社区驱动的编排策略、新连接器以及最佳实践。

This means Microsoft Agent Framework is not a fixed product — it’s a living ecosystem, continuously shaped by contributions from Microsoft Research and the global OSS community.

这意味着 Microsoft Agent Framework 并非一成不变的产品，而是一个不断发展的生态系统，持续受到微软研究院和全球开源社区贡献的塑造。

4. Ready for Production    生产环境就绪

Microsoft Agent Framework isn’t just for experimentation — it was built for enterprise-grade deployment from the very beginning. It delivers the end-to-end tooling and runtime features needed to confidently move from prototype to scale, while integrating deeply with the Azure AI Foundry ecosystem.

Microsoft Agent Framework 不仅仅适用于实验探索——它从一开始就是为满足企业级部署需求而构建的。它提供了端到端的工具链和运行时功能，让您能够自信地将原型转化为规模化部署，同时与 Azure AI Foundry 生态系统深度集成。

• Observability: OpenTelemetry can instrument and visualize every agent action, tool invocation, and orchestration step, making it easy to trace reasoning flows and monitor performance through Azure AI Foundry dashboards.    可观察性：通过 OpenTelemetry，可以对每个智能体的操作、工具调用和编排步骤进行监控和可视化，便于通过 Azure AI Foundry 仪表板追踪推理流程并监控性能。
• Secure Cloud Hosting: Agents will run natively on Azure AI Foundry with enterprise controls like virtual network integration, role-based access, private data handling, and built-in content safety.    安全的云托管：智能体将在 Azure AI Foundry 上原生运行，具备虚拟网络集成、基于角色的访问控制、私有数据处理和内置内容安全等企业级管控功能。
• Security & compliance: Azure AI Content Safety integration, Entra ID authentication, and structured logging mean Agent Framework agents can run in regulated industries.    安全与合规：集成 Azure AI 内容安全服务、Entra ID 身份验证和结构化日志记录，意味着 Agent Framework 的智能体能够在受监管的行业中运行。
• Long-running durability: Agent threads and workflows can pause, resume, and recover from interruptions, with retry and error-handling logic ensuring long-running processes remain reliable at scale.    长时间运行的持久性：智能体的会话和工作流可以暂停、恢复并从中断中恢复，配合重试机制和错误处理逻辑，确保长时间运行的流程在大规模下依然可靠。
• Human in the loop: For scenarios that require governance, tools can be marked as requiring human approval. Agent Framework automatically emits a pending approval request that can be routed to a UI or queue, then continues or denies execution accordingly. This works across local tools or remote service calls, ensuring sensitive operations remain under control.    人工介入流程：在需要监管的场景中，可将工具标记为需经人工审批。Agent Framework会自动发出待审批请求，该请求可被路由至用户界面或队列，进而根据审批结果继续执行或予以拒绝。此机制对本地工具和远程服务调用均适用，确保敏感操作始终处于受控状态。
• CI/CD integration: The framework integrates directly into GitHub Actions and Azure DevOps pipelines, with telemetry flowing into Azure Monitor and Application Insights for enterprise-grade deployment and root-cause analysis.    CI/CD 集成：该框架可直接集成到 GitHub Actions 和 Azure DevOps 流水线中，其遥测数据流入 Azure Monitor 和 Application Insights，为企业级部署与根因分析提供支持。

With these capabilities, Microsoft Agent Framework makes it seamless to prototype locally, debug with rich telemetry, and then scale securely into production with the enterprise readiness that modern AI systems demand.

凭借这些能力，Microsoft Agent Framework 让您能够轻松地在本地进行原型开发，利用丰富的遥测数据进行调试，然后安全、顺畅地扩展至生产环境，满足现代 AI 系统对企业级就绪性的严苛要求。

Customer Momentum    客户应用态势

Enterprises across industries are already testing Microsoft Agent Framework in real-world scenarios:

各行各业的企业已在真实场景中试用 Microsoft Agent Framework：

• KPMG is building Clara AI, a multi-agent system that automates audit testing and documentation. “Foundry Agent Service and Microsoft Agent Framework connect our agents to data and each other, and the governance and observability in Azure AI Foundry provide what KPMG firms need to be successful in a regulated industry” – Sebastian Stöckle, Global Head of Audit Innovation and AI, KPMG International    毕马威 (KPMG) 正在构建 Clara AI，一个自动化审计测试和文档工作的多智能体系统。“Foundry Agent Service 和 Microsoft Agent Framework 将我们的智能体相互连接，并接入数据，而 Azure AI Foundry 中的治理和可观察性功能，正是毕马威各成员所处受监管行业取得成功所需的关键。”——毕马威国际审计创新与人工智能全球主管 Sebastian Stöckle
• Commerzbank is piloting Microsoft Agent Framework to power avatar-driven customer support, enabling more natural, accessible, and compliant customer interactions. “The new Microsoft Agent Framework simplifies coding, reduces efforts and fully supports MCP for agentic solutions. We are really looking forward to the productive usage of container-based Azure AI Foundry agents, which significantly reduces workload in IT operations” – Gerald Ertl, Managing Director/Head of Digital Banking Solutions, Commerzbank AG    德国商业银行正在试点使用 Microsoft Agent Framework 来驱动由虚拟形象（avatar）提供支持的客户服务，从而实现更自然、更便捷且符合合规要求的客户互动。“全新的 Microsoft Agent Framework 简化了编码，减少了工作量，并全面支持 MCP（模型上下文协议）以构建智能体解决方案。我们非常期待容器化 Azure AI Foundry 智能体的生产应用，这将显著减轻 IT 运维的工作负担。”——德国商业银行数字银行解决方案董事总经理/主管 Gerald Ertl
• BMW: BMW is using Microsoft Agent Framework and Foundry Agent Service to orchestrate multi-agent systems that analyze terabytes of vehicle telemetry in near real time, enabling engineers to accelerate design cycles and spot issues earlier in testing. “Durability and observability are key for our operations. With multi-agent systems powered by Microsoft Agent Framework and Foundry Agent Service, engineers don’t just access data — they get insights they can act on immediately, cutting analysis from days to minutes.” – Christof Gebhart, Manager, Advanced Vehicle Measurement Technology, BMW    宝马：宝马正在使用 Microsoft Agent Framework 和 Foundry Agent Service 来编排多智能体系统，近乎实时地分析兆字节级别的车辆遥测数据，使工程师能够加速设计周期，并在测试早期发现潜在问题。“持久性和可观察性对我们的运营至关重要。借助由 Microsoft Agent Framework 和 Foundry Agent Service 驱动的多智能体系统，工程师不仅能获取数据，更能获得可立即指导行动的深度洞察，将分析时间从数天压缩至分钟级。”——宝马高级车辆测量技术经理 Christof Gebhart
• Fujitsu is embedding Microsoft Agent Framework into its integration services, enabling customers to safely adopt advanced orchestration strategies such as group chat and debate. “I believe it enables us to build multi-agent systems that emphasize coexistence between humans and AI and can truly accelerate our AI transformation” – Hirotaka Ito, Lead engineer of AI, Corporate Digital Unit, Fujitsu.    富士通  正将 Microsoft Agent Framework 嵌入其集成服务中，使客户能够安全地采用诸如群组对话和辩论等先进的编排策略。“我相信这能帮助我们构建强调人与人工智能共存的多智能体系统，真正加速我们的 AI 转型。”——富士通公司数字事业部 AI 首席工程师 伊藤博隆 (Hirotaka Ito)
• Citrix: Citrix is exploring how they can use agentic AI within VDI environments to improve enterprise productivity and efficiency. “We are excited about the Microsoft Agent Framework, which brings a modern, developer-first approach to building single- and multi-agent workflows. With support of key APIs and languages, and native adoption of emerging protocols for tool calling and observability, it enables intuitive development of agents on Azure AI Foundry, without compromising developer control. We are eager to leverage the framework to deliver enterprise-scale, production-ready AI solutions to our customers.” — George Tsolis, Distinguished Engineer, Citrix    思杰：Citrix 正在探索如何在 VDI 环境中应用智能体 AI，以提升企业生产力和效率。“我们对 Microsoft Agent Framework 感到非常兴奋，它为构建单智能体和多智能体工作流带来了现代化的、以开发者为先的方法。凭借对关键 API 和语言的支持，以及对新兴工具调用和可观察性协议的原生采用，它使得在 Azure AI Foundry 上进行智能体开发变得直观，同时不牺牲开发者的控制权。我们热切期待利用该框架，为我们的客户提供企业级、可投入生产的 AI 解决方案。”——Citrix 杰出工程师 George Tsolis
• Fractal: Fractal’s Cogentiq is an agentic AI platform that uses Microsoft Agent Framework to orchestrate and scale enterprise AI agents and workflows across industries. “Cogentiq leverages Microsoft Agent Framework to orchestrate and scale AI agents and workflows. Microsoft Agent framework’s ease of use, flexible agent development and deployment options and support for building complex multi-agentic workflows enables us to rapidly build, deploy, and manage multi-agent solutions across industries and functions. The Agent Framework allows both technical and business teams to innovate quickly, integrate with enterprise systems, and deliver value at scale through production-ready tools and access to industry-leading AI models. It takes care of the heavy lifting around model access, deployment, scaling, security, networking helping Fractal focus on solving our client’s industry and function specific business problems.” — Himanshu Nautiyal, Chief Product Officer, Fractal    Fractal：Fractal 的 Cogentiq 是一个智能体 AI 平台，利用 Microsoft Agent Framework 在不同行业间编排和扩展企业级 AI 智能体及工作流。“Cogentiq 利用 Microsoft Agent Framework 来编排和扩展 AI 智能体及工作流。Microsoft Agent Framework 易于使用、提供灵活的智能体开发与部署选项，并支持构建复杂的多智能体工作流，使我们能够跨行业和职能快速构建、部署和管理多智能体解决方案。该框架让技术和业务团队都能快速创新，与企业系统集成，并通过生产就绪的工具和对行业领先 AI 模型的访问，规模化交付价值。它承担了模型访问、部署、扩展、安全和网络等方面的繁重工作，让 Fractal 能够专注于解决客户特定行业和职能的业务问题。”——Fractal 首席产品官 Himanshu Nautiyal
• TCS: Tata Consultancy Services is actively building a multi-agent practice on the Microsoft Agent Framework, with several initiatives underway that showcase their strategic investment and technical depth including agentic solutions for finance, IT operations, and retail. “Adopting Microsoft Agent Framework is not just a technological advancement, but a bold step towards reimagining industry value chains. By harnessing Agentic AI and Frontier models, we enable our teams to build flexible, scalable, enterprise-grade solutions that transform workflows and deliver value across platforms. True leadership is about empowering innovation, embracing change, and fostering an environment where agility and collaboration drive excellence.”  – Girish Phadke, Head, Microsoft Azure Practice, TCS    塔塔咨询服务：塔塔咨询服务正在 Microsoft Agent Framework 上积极构建多智能体实践，多个正在进行的项目展示了其战略投入和技术深度，包括为金融、IT 运维和零售领域打造的智能体解决方案。“采用 Microsoft Agent Framework 不仅是一项技术进步，更是迈向重塑行业价值链的大胆一步。通过利用智能体 AI 和前沿模型，我们使团队能够构建灵活、可扩展、企业级的解决方案，变革工作流程，并跨平台交付价值。真正的领导力在于赋能创新、拥抱变革，并营造一个敏捷与协作驱动卓越的环境。”——TCS 微软 Azure 实践主管 Girish Phadke
• Sitecore: Sitecore is building a solution to help marketers interact more seamlessly with the Sitecore platform by automating tasks across content supply chain, from creating and managing web experiences to digital assets, using intelligent agents. “By partnering with Microsoft to leverage its new Microsoft Agent Framework, Sitecore can bring together the best of both worlds: the flexibility to power fully non-deterministic agentic orchestrations and the reliability to run more deterministic, repeatable agents. At the same time, we benefit from Microsoft’s enterprise-grade observability and telemetry, ensuring that these orchestrations are not only powerful but also secure, measurable, and production-ready.” – Mo Cherif, VP of AI, Sitecore.    Sitecore：Sitecore 正在构建一个解决方案，通过智能体自动化内容供应链中的任务（从创建和管理网络体验到数字资产），帮助营销人员更无缝地与 Sitecore 平台交互。“通过与微软合作并利用其全新的 Microsoft Agent Framework，Sitecore 能够融合两者的最佳优势：既能实现完全非确定性的智能体编排的灵活性，又能运行更具确定性、可重复的智能体以确保可靠性。同时，我们受益于微软的企业级可观察性和遥测能力，确保这些编排不仅强大，而且安全、可衡量且可投入生产。”——Sitecore 人工智能副总裁 Mo Cherif
• NTT DATA: Agentic AI value includes a complete ecosystem of solutions, services, and partners. NTT DATA is adopting the Microsoft Agent Framework in alignment with efforts to standardize its R&D approach for multi-agent management, enabling the company to deploy, manage, and optimize AI solutions across industries. This will help accelerate deployments, support complex process workflows that can be customized and replicated and make it easier to connect and orchestrate sophisticated models on behalf of clients. “By adopting the Microsoft Agent Framework, NTT DATA is not only further standardizing how we develop and manage multi-agent systems, but also accelerating how our clients realize value from AI. This initiative allows us to deliver faster, more scalable, and more governed AI solutions, while staying closely in step with Microsoft’s engineering roadmap.”- Charlie Doubek, Global VP, Agentic AI Services Leader, Cloud and Security, NTT DATA    NTT DATA：智能体 AI 价值包含一个完整的解决方案、服务和合作伙伴生态系统。NTT DATA 正在采用 Microsoft Agent Framework，以配合其标准化多智能体管理研发方法的努力，使公司能够跨行业部署、管理和优化 AI 解决方案。这将有助于加速部署，支持可定制和复制的复杂流程工作流，并简化为客户连接和编排复杂模型的过程。“通过采用 Microsoft Agent Framework，NTT DATA 不仅进一步标准化了我们开发和管理多智能体系统的方式，也加速了客户从 AI 中获取价值的速度。这一举措使我们能够交付更快、更具可扩展性且更受管控的 AI 解决方案，同时与微软的工程路线图保持紧密同步。”——NTT DATA 全球副总裁、云与安全智能体 AI 服务负责人 Charlie Doubek
• MTech Systems: MTech Systems will use the new Agent Framework to orchestrate transactional data anomaly sweeps, human-in-the-loop approvals, and automated fixes – agent patterns that previously required extensive glue code. “The framework gives us a batteries-included developer experience and makes agent workflows far easier to build and run. Features like checkpointing and declarative YAML workflows will save us time and let us scale changes across hundreds of customer applications without redeploys” – Barry Schulz, CTO, MTech Systems    MTech Systems：MTech Systems 将利用新的 Agent Framework 来编排事务性数据异常扫描、人工审批环节和自动修复——这些智能体模式以往需要大量胶水代码。“该框架为我们提供了开箱即用的开发者体验，使智能体工作流的构建和运行变得容易得多。检查点（checkpointing）和声明式 YAML 工作流等功能将为我们节省时间，并让我们无需重新部署，即可在数百个客户应用中规模化地实施变更。”——MTech Systems 首席技术官 Barry Schulz
• TeamViewer: TeamViewer is embedding agentic AI into its IT support stack so that remote support agents can get real-time diagnostics, automated summarization, and contextual recommendations during sessions. “The framework strikes the right balance between technical depth and usability. Its intuitive design and modular structure make it easy for our teams to adopt quickly, while providing the scalability and flexibility we need for complex projects. That combination allows us to deliver value faster today and positions us well to take advantage of the enhancements still to come.”– Mei Dent, Chief Product and Technology Officer TeamViewer    TeamViewer：TeamViewer 正将智能体 AI 嵌入其 IT 支持堆栈中，以便远程支持人员在会话期间获得实时诊断、自动摘要和情境化建议。“该框架在技术深度和易用性之间取得了恰到好处的平衡。其直观的设计和模块化结构使我们的团队能够快速采用，同时为我们复杂项目所需的可扩展性和灵活性提供了保障。这种结合使我们能够更快地交付价值，并为利用未来即将到来的增强功能做好了充分准备。”——TeamViewer 首席产品与技术官 Mei Dent
• Weights & Biases: Weights & Biases is collaborating with Microsoft to ensure developers can seamlessly train, track, and operationalize AI agents at scale. “The new Microsoft Agentic Framework makes building production-ready agents dramatically easier. With flexible orchestration, checkpointing to save time and compute, and built-in human-in-the-loop support, it tackles the real challenges teams face when moving from prototype to production “ – Phil Gurbacki, VP of Product, Weights & Biases    Weights & Biases：Weights & Biases 正与微软合作，确保开发者能够无缝地大规模训练、跟踪和运维 AI 智能体。“全新的 Microsoft 智能体框架使构建可投入生产的智能体变得极为简单。凭借灵活的编排、用于节省时间和算力的检查点功能，以及内置的人工介入支持支持，它解决了团队从原型走向生产时面临的真实挑战。”——Weights & Biases 产品副总裁 Phil Gurbacki
• Elastic: Elasticsearch supports a native connector to Microsoft Agent Framework, enabling developers to seamlessly integrate enterprise data into intelligent agents and workflows. “Elasticsearch is the context engineering platform and vector database of choice for organizations to store and search their most valuable operational and business data. With the new Microsoft Agent Framework connector, developers can now bring the most relevant organizational context directly into intelligent agents and multi-agent workflows. This makes it easier than ever to build production-ready AI solutions that combine the reasoning power of agents with the speed and scale of Elasticsearch.” — Steve Kearns, General Manager Search Solutions, Elastic    Elastic：旗下Elasticsearch为Microsoft Agent Framework提供原生连接器，助力开发者将企业数据无缝集成至智能体与工作流。“Elasticsearch 是组织存储检索核心运营与业务数据的首选上下文工程平台和向量数据库。借助新的 Microsoft Agent Framework 连接器，开发者现在可以将最相关的组织上下文直接引入智能体和多智能体工作流中。这使得构建结合了智能体推理能力与 Elasticsearch 速度和规模的生产就绪 AI 解决方案变得前所未有的简单。”——Elastic 搜索解决方案总经理 Steve Kearns

These early stories highlight the dual promise of Microsoft Agent Framework: innovative enough to inspire new approaches, stable enough to deploy in production.

这些早期案例凸显了 Microsoft Agent Framework 的双重承诺：足够创新，能够激发新的方法；足够稳定，能够投入生产环境使用。

Path to Microsoft Agent Framework    迈向Microsoft Agent Framework之路

Many customers are already using Semantic Kernel or AutoGen in production today. Both projects will remain supported but most investment is now focused on Microsoft Agent Framework. Developers using Semantic Kernel or AutoGen will find the transition straightforward:

如今，许多客户已在生产环境中使用 Semantic Kernel 或 AutoGen。这两个项目将继续获得支持，但目前大部分投入已集中于 Microsoft Agent Framework。使用 Semantic Kernel 或 AutoGen 的开发者会发现迁移过程十分顺畅：

• For Semantic Kernel users:    对于 Semantic Kernel 用户：
  • Migration is straightforward: replace Kernel and plugin patterns with the Agent and Tool abstractions.    迁移过程简单明了：将 Kernel 和插件模式替换为 Agent（智能体）和 Tool（工具）的抽象。
  • .NET developers move from Microsoft.SemanticKernel.* to the new Microsoft.Extensions.AI.* namespaces, with agents created directly from providers instead of requiring Kernel coupling.    .NET 开发者将从 Microsoft.SemanticKernel.* 迁移到新的 Microsoft.Extensions.AI.* 命名空间。智能体现在直接由提供程序创建，不再需要与 Kernel 耦合。
  • Python developers can install the full package (pip install agent-framework) or just the components they need (e.g., agent-framework-azure-ai, agent-framework-redis).    Python 开发者可以安装完整包（pip install agent-framework），或仅安装所需组件（例如 agent-framework-azure-ai、agent-framework-redis）。
  • Agents now manage threads natively, simplify invocation with RunAsync / RunStreamingAsync, and register tools inline without attributes or plugin wrappers.    智能体现在原生管理会话，通过 RunAsync / RunStreamingAsync 简化调用，并以内联方式注册工具，无需特性或插件包装器。
  • Existing vector store integrations (Azure AI Search, Postgres, Cosmos DB, Redis, Elasticsearch, etc.) continue to work through connectors.    现有的向量存储集成（Azure AI Search、PostgreSQL、Cosmos DB、Redis、Elasticsearch 等）可通过连接器继续使用。
  • Plugins like Bing, Google, OpenAPI, and Microsoft Graph port directly as tools, often exposed via MCP or OpenAPI.    Bing、Google、OpenAPI 及 Microsoft Graph 等插件可直接移植为工具，通常通过 MCP 或 OpenAPI 暴露。
  • The net result: less boilerplate, simplified memory management, and alignment with open standards.    最终效果是：样板代码更少，内存管理更简化，并与开放标准保持一致。
• For AutoGen users:    对于 AutoGen 用户：
  • AutoGen pioneered many orchestration patterns (GroupChat, GraphFlow, event-driven runtimes), which are now unified in Agent Framework under the Workflow abstraction.    AutoGen 开创了许多编排模式（如 GroupChat、GraphFlow、事件驱动运行时），这些模式现在已统一在 Agent Framework 的 Workflow（工作流） 抽象之下。
  • The AssistantAgent maps directly to the new ChatAgent, which is multi-turn by default and continues tool invocation until a result is ready.    AssistantAgent 直接对应到新的 ChatAgent，后者默认支持多轮对话，并持续调用工具直至结果就绪。
  • FunctionTool wrappers migrate to the @ai_function decorator, with automatic schema inference and support for hosted tools like code interpreter or web search.    FunctionTool 包装器迁移到 @ai_function 装饰器，支持自动模式推断，并兼容代码解释器或网络搜索等托管工具。
  • Messaging is simplified: multiple message classes are replaced with a unified ChatMessage type, with explicit roles (USER, ASSISTANT, TOOL, SYSTEM).    消息机制得到简化：多种消息类被统一的 ChatMessage 类型取代，并明确区分角色（USER、ASSISTANT、TOOL、SYSTEM）。
  • Orchestration shifts from event-driven models to a typed, graph-based Workflow API that supports checkpointing, pause/resume, and human-in-the-loop flows.    编排方式从事件驱动模型转变为类型化的、基于图的 Workflow API，支持检查点（checkpointing）、暂停/恢复以及人工介入流程。
  • Observability is richer and simpler, with OpenTelemetry support out of the box.    可观察性更丰富且更简单，开箱即用支持 OpenTelemetry。
  • Most single-agent migrations require only light refactoring; multi-agent migrations benefit from the new Workflow model with stronger composability and durability.    大多数单智能体迁移仅需少量重构；多智能体迁移则受益于新的 Workflow 模型，具备更强的可组合性和持久性。

This continuity means developers can preserve their existing investments while unlocking new capabilities. Microsoft Agent Framework is not a replacement for what came before — it is the natural evolution that unites innovation and stability. For more information about migration, see the documentation.

这种延续性意味着开发者可以在保留现有投资的同时，解锁全新的能力。Microsoft Agent Framework 并非对过往技术的简单替代，而是融合了创新与稳定的自然演进。如需了解有关迁移的更多信息，请参阅相关文档。

Get Started with Microsoft Agent Framework Today    立即开始使用 Microsoft Agent Framework

Agents are fast becoming the next layer of application logic — reasoning about goals, calling tools, collaborating with each other, and adapting dynamically. With Microsoft Agent Framework, developers now have a single, open-source foundation that carries the best of research innovation into production with the durability, observability, and enterprise readiness required to scale.

智能体正迅速成为应用逻辑的新层级——它们能够推理目标、调用工具、相互协作，并动态匹配。借助 Microsoft Agent Framework，开发者现在拥有一个统一的开源基础，能够将前沿研究成果的最佳实践带入生产环境，并具备规模化所需的持久性、可观察性和企业级就绪能力。

This is the natural evolution of the journey that began with Semantic Kernel and AutoGen — and it’s only the beginning. By building in the open and co-creating with the developer community, Microsoft Agent Framework will continue to evolve as the foundation for next-generation multi-agent systems.

这是从Semantic Kernel和AutoGen启程之旅的自然演进——而这仅仅只是开始。通过开放构建并与开发者社区共同创造，Microsoft Agent Framework将持续发展，成为下一代多智能体系统的基石。

• • Download the SDK: aka.ms/AgentFramework    下载SDK：aka.ms/AgentFramework
  • Explore documentation for more details: https://aka.ms/AgentFramework/Docs    探索文档：https://aka.ms/AgentFramework/Docs
  • See it in action: Watch demos on AI Show and Open at Microsoft    观看演示：在 AI Show 和 Open at Microsoft 观看实际演示
  • Learn step by step: Microsoft Learn modules for Agent Framework and AI Agents for Beginners    循序渐进学习：通过 Microsoft Learn 的 Agent Framework 以及《AI Agents for Beginners》课程学习
  • Join the Azure AI Foundry Discord to connect with developers and product groups, sharpen your AI skills, and stay inspired through real-time community. Join us for an AMA Tuesday 7^th October 9am PST: https://aka.ms/foundry/discord加入社区：加入 Azure AI Foundry Discord 社区，与开发者和产品团队交流，精进您的 AI 技能，并通过实时社区互动获得灵感。加入我们于 10 月 7 日（星期二）太平洋时间上午 9 点举行的问答会（AMA）：https://aka.ms/foundry/discord