Semantic Kernel（SK）Semantic Kernel（SK）
Semantic Kernel 是微软的企业级 AI 嵌入 SDK（MIT，C#/.NET + Python 双线）： connector 接模型、skills/plugins 挂函数，把 LLM 嵌进现有应用进程。1.x 线仍在发版 （python-1.46.0，2026-10-28）；但官方已指明 Microsoft Agent Framework 为其指定后继—— 现役维护与新投资重心是两条线。
Semantic Kernel is Microsoft's enterprise AI-embedding SDK (MIT, C#/.NET + Python both releasing): connectors for models, skills/plugins for functions. The 1.x line still ships — but the official docs name Microsoft Agent Framework as its designated successor.
适合与不适合When to use it, when not to
适合：.NET/企业栈嵌入 AI、Azure 生态、要双语言 SDK。 不适合：新项目起步（先看 MAF）。
Fits: .NET/enterprise stacks embedding AI into existing apps, the Azure ecosystem, dual-language SDKs. Not a fit: greenfield starts (look at MAF first).
中文生态China availability 国产 / 国内可直接用 / 备案状态 · 快照domestic / direct access / CAC filing · snapshot 2026-10-31
- 出品属地Origin
- 国外出品foreign-made Microsoft（美国）；learn.microsoft.com 大陆可达性良好。Microsoft (US); learn.microsoft.com reaches well from the mainland.
- 国内使用Direct access
- 大陆可直连direct OK learn.microsoft.com/en-us/semantic-kernel 直连实测 200（2026-10-31）；GitHub 仓大陆直连波动（同日代理可达）。learn.microsoft.com/en-us/semantic-kernel measured 200 direct (2026-10-31); the GitHub repo's mainland reachability fluctuates (200 via proxy the same day).
- 备案状态CAC filing
- 未见于备案名单not in filing list：未见于网信办公示名单（2026-10-31 检索）；开源 SDK 不面向大陆公众提供服务，备案不适用（Azure 中国区服务属另一主体线，未核验不下断言）。not in the CAC public registry (searched 2026-10-31); an open-source SDK not serving the mainland public, so filing does not apply (Azure China services belong to another entity line — not verified, no assertion made).
备案信息以网信办《生成式人工智能服务已备案信息》公示名单为准（按批次滚动更新，本站只抄编号不解读）；「大陆可直连」是本站从大陆网络直连实测的结果，不是厂商承诺。GitHub 类站点当日实测有波动，口径写在依据里。Filing status is quoted from the CAC public registry (updated batch by batch; this site copies the number and does not interpret it). "Direct access" is this site's own reachability measurement from a mainland network, not a vendor promise. GitHub-hosted sites fluctuated on the measurement day — the note says so.
固定坐标系Fixed coordinate system
8 个维度dimensions, 与同分区其他对象逐项可比each comparable item by item against others in the same section.
- 模型与开放条件Model & openness
connector 式多模型接入（OpenAI/Azure OpenAI/Anthropic 等，未逐条清点）； 定位是把 LLM 能力嵌进企业应用（官方仓描述「Integrate cutting-edge LLM technology quickly and easily into your apps」）。
Connector-style multi-model access (OpenAI / Azure OpenAI / Anthropic etc., not itemized); the positioning is embedding LLM capability into enterprise applications (official repo description: "Integrate cutting-edge LLM technology quickly and easily into your apps").
- 运行位置Where it runs
多语言 SDK 形态（C#/.NET 与 Python 双线发版——dotnet-1.81.0 与 python-1.46.0 均在 2026-10 发版， GitHub API 核验）：库嵌进你的应用进程，不是独立运行时。GitHub microsoft/semantic-kernel 28,724★、最近推送 2026-10-30（API 核验，仍活跃发版）。
A multi-language SDK form (C#/.NET and Python lines both release — dotnet-1.81.0 and python-1.46.0 both shipped in 2026-10, GitHub API verified): a library embedded in your application process, not a standalone runtime. GitHub microsoft/semantic-kernel 28,724 stars, last push 2026-10-30 (API verified, still releasing actively).
- 本地文件Local files
库形态嵌在你应用里，数据面随你的应用；Azure 侧服务（Azure OpenAI 等）数据在云侧——按你接的 connector 分。
As a library it lives inside your app, so the data surface follows your app; Azure-side services (Azure OpenAI etc.) keep data cloud-side — it splits by the connector you wire in.
- 关机后的任务Tasks after shutdown
进程内调用为主（库形态）；常驻语义随宿主应用，SK 本身不提供平面。
In-process calls are the main mode (library form); residency semantics follow the host application — SK itself provides no plane.
- 工具与扩展Tools & extensions
skills/plugins 函数调用线是老牌卖点（SK 早期以「kernel + skills」抽象出名）；MCP 支持状态未逐条核验。
The skills/plugins function-calling line is the long-standing selling point (SK early on became known for the "kernel + skills" abstraction); MCP support status not itemized.
- 上下文与记忆Context & memory
企业应用嵌入场景的上下文管理（planner/agent 线细节未逐条核验）。
Context management for enterprise-app embedding scenarios (planner/agent line details not itemized).
- 权限与限制Permissions & limits
企业选型面是主轴（Azure 生态鉴权/合规线随微软云体系）； 库本身权限面随宿主应用。
The enterprise selection surface is the main axis (Azure-ecosystem auth/compliance lines follow the Microsoft cloud system); the library's own permission surface follows the host application.
- 适合什么任务Best for
适合：.NET/企业栈要把 AI 嵌进现有应用、Azure 生态、要双语言（C#/Python）SDK。 不适合：新项目起步——官方已指明 Microsoft Agent Framework 为其后继（见 pitfalls）。
Fits: .NET/enterprise stacks embedding AI into existing apps, the Azure ecosystem, needing dual-language (C#/Python) SDKs. Does not fit: greenfield starts — the official docs name Microsoft Agent Framework as its successor (see pitfalls).
头号误解Top misconception
- 后继关系要写清：官方文档称 Microsoft Agent Framework（MAF）为 SK+AutoGen 的指定后继——SK 仍发版维护（1.x 线 2026-10 仍在出）但新投资重心在 MAF；新项目选型先看 MAF 档案State the successor relation clearly: the official docs name Microsoft Agent Framework (MAF) as the designated successor of SK + AutoGen — SK still releases maintenance (the 1.x line still ships in 2026-10) but new investment centers on MAF; new-project selections should read the MAF profile first
- 同厂三件别混：SK（现役企业 SDK）/ AutoGen（研究线，不收）/ MAF（后继）——微软 agent 框架线三段，本站收 SK 与 MAF 两份Do not mix three same-vendor pieces: SK (active enterprise SDK) / AutoGen (research line, not collected) / MAF (successor) — three stages of Microsoft's agent-framework line; this site collects SK and MAF
- 双语言发版线独立：python-1.46.0 与 dotnet-1.81.0 版本号不同步——按语言线分别看版本Dual-language release lines are independent: python-1.46.0 and dotnet-1.81.0 version numbers are not in sync — read versions per language line
未知项清单Known unknowns
- MCP 支持状态（未逐条核验）
- SK → MAF 的官方迁移指南覆盖度（未逐条核对）
- planner/agent 线各组件的维护状态细分（未逐条核验）
- MCP support status (not itemized)
- Coverage of the official SK → MAF migration guide (not checked item by item)
- Maintenance-status breakdown of individual planner/agent components (not itemized)
证据来源Evidence sources
判断可回到以下一手源复核。本站核验日Every judgement can be rechecked against the primary sources below. Verified on 2026-10-31，内容更新日content updated on 2026-10-31。
| 类型Type | 名称Name | 链接Link |
|---|---|---|
| repo | microsoft/semantic-kernel · 仓库（MIT；28,724★，最近推送 2026-10-30，GitHub API 核验）microsoft/semantic-kernel · repository (MIT; 28,724 stars, last push 2026-10-30, GitHub API verified) | https://github.com/microsoft/semantic-kernel |
| docs | Semantic Kernel · 官方文档Semantic Kernel · official docs | https://learn.microsoft.com/en-us/semantic-kernel/ |
| changelog | microsoft/semantic-kernel · Releases（python-1.46.0 2026-10-28 / dotnet-1.81.0 2026-10-06）microsoft/semantic-kernel · Releases (python-1.46.0 2026-10-28 / dotnet-1.81.0 2026-10-06) | https://github.com/microsoft/semantic-kernel/releases |
| docs | Microsoft Agent Framework · 官方文档（SK 的官方指定后继，迁移参照）Microsoft Agent Framework · official docs (the official designated successor of SK, migration reference) | https://learn.microsoft.com/en-us/agent-framework/ |
实测记录Measured runs
本档未跑集成与性能实测；结论只覆盖 API 与官方文档披露的能力面。
This site has not completed measured runs. The test protocol is at protocols/framework/_protocol.md.
相关条目Related entries
- Microsoft Agent FrameworkMicrosoft Agent Framework同大类对照：同厂指定后继（SK+AutoGen 合流线）——SK 现役维护，MAF 承接新投资，选型按项目阶段分。Same-category contrast: the same-vendor designated successor (the SK+AutoGen convergence line) — SK stays in active maintenance, MAF takes the new investment; choose by project stage.
- AgnoAgno (formerly Phidata)同大类对照：Agno 自带控制平面（AgentOS），SK 只做库嵌你应用——「给平面」与「给库」两种企业形态。Same-category contrast: Agno ships its own control plane (AgentOS); SK is only a library embedded in your app — "give a plane" vs. "give a library", two enterprise forms.
- Microsoft CopilotMicrosoft Copilot跨大类对照：同为微软 AI 线，Copilot 是成品助手，SK 是让你把 AI 嵌进自己应用的底座——成品与底座两面。Cross-category contrast: both are Microsoft AI lines — Copilot is a finished assistant, SK is the base for embedding AI into your own apps — product vs. base, two faces.
本页由This page is generated from the ai-handbook 数据层生成（CC BY 4.0）。data layer (CC BY 4.0).
方法论与坐标系定义见仓库内For the methodology and coordinate definitions, see METHODOLOGY.md.