Microsoft Semantic Kernel Documentation
by Microsoft
Microsoft's agent SDK documented for C#, Python and Java — with a dependency-injection kernel at the centre.
Overview
The docs describe Semantic Kernel as "a lightweight, open-source development kit that lets you easily build AI agents and integrate the latest AI models into your C#, Python, or Java codebase", positioning it as middleware rather than an application framework. The central concept is the kernel, documented as a dependency-injection container holding two component types — services (chat completion, logging, HTTP clients) and plugins (native code, prompt templates, OpenAPI specifications, MCP tools). When you invoke a prompt the kernel selects the AI service, builds the prompt from its template, sends it, parses the response and returns it, with hooks, filters and middleware available at every step for logging, telemetry and responsible-AI enforcement. The Agent Framework section documents ChatCompletionAgent, OpenAIAssistantAgent, AzureAIAgent and OpenAIResponsesAgent plus a separate orchestration package, with installation tables per language: NuGet packages for .NET, the semantic-kernel PyPI module for Python, and Maven artifacts for Java. Python 3.10+, .NET 10.0+ and Java 17+ are supported. Newer pages document exposing kernel functions as an MCP server via kernel.as_mcp_server() over stdio or SSE, and surfacing prompt templates as MCP prompts. Microsoft states v1.0+ across all three languages with a commitment to non-breaking changes. The repository has roughly 28,400 stars under MIT, and now points to Microsoft Agent Framework — the convergence of Semantic Kernel and AutoGen — as its enterprise successor; a Microsoft maintainer has stated on the repository that "There are no plans to discontinue Semantic Kernel", so read these docs knowing new agent work is being directed at the newer framework.
At a Glance
- Topic
- Frameworks
- Level
- Intermediate
- Format
- Documentation
- Cost
- Free
- Duration
- Reference documentation; the quick start plus the agent and process framework guides take roughly 3-5 hours
- Provider
- Microsoft
- Hands-on
- Yes — code/exercises
- Certificate
- None
What You’ll Learn
- ✓Build a kernel and register AI services and plugins through dependency injection
- ✓Expose existing code and APIs to a model as callable plugins and function tools
- ✓Use hooks, filters and telemetry to log, monitor and govern agent behaviour centrally
- ✓Create agents with ChatCompletionAgent, OpenAIAssistantAgent, AzureAIAgent or OpenAIResponsesAgent
- ✓Orchestrate multiple agents together using the dedicated agent orchestration package
- ✓Turn a Semantic Kernel instance into an MCP server over stdio or SSE transports
- ✓Connect vector stores and local runtimes such as Azure AI Search, Elasticsearch, Chroma, Ollama and ONNX
Highlights
- •One of very few agent frameworks with first-class C#, Python AND Java documentation, code samples pivoted per language
- •Enterprise concerns are documented rather than assumed: telemetry, hooks, filters and a stated non-breaking-change commitment on v1.0+
- •MCP support runs both directions — consume MCP tools as plugins, or expose your kernel as an MCP server
- •Backed by an MIT-licensed repository with ~28.4k stars and ~5,051 commits, so the docs track shipped code
- •Plugins use OpenAPI specifications, the same mechanism as Microsoft 365 Copilot extensions
- •Worth knowing before you commit: Microsoft now positions Agent Framework as the successor, while stating publicly that Semantic Kernel is not being discontinued
Who It’s For
Best For
- ✓Enterprise .NET and Java teams adding AI agents to an existing application
- ✓Architects who need model-agnostic middleware rather than a single vendor's SDK
- ✓Developers building agents that must call existing internal APIs and services
- ✓Teams with governance, telemetry or responsible-AI review requirements
Prerequisites
- •Working knowledge of C#, Python or Java — the docs assume you are integrating into an existing codebase
- •Familiarity with dependency injection and service-container patterns, which the kernel is modelled on
- •An OpenAI, Azure OpenAI or other supported model endpoint to run the samples
FAQ
What is Microsoft Semantic Kernel Documentation?
Semantic Kernel is Microsoft's open-source development kit for building AI agents and integrating models into existing C#, Python or Java codebases, and this is its official documentation. It covers the kernel itself, plugins, agents, orchestration and the process framework, with every code sample presented per language. It is the reference to read if you are adding agents to an existing enterprise application rather than starting greenfield.
Is Microsoft Semantic Kernel Documentation free?
Microsoft Semantic Kernel Documentation is free to access.
What level is Microsoft Semantic Kernel Documentation for?
Microsoft Semantic Kernel Documentation is aimed at a intermediate audience. Recommended background: Working knowledge of C#, Python or Java — the docs assume you are integrating into an existing codebase, Familiarity with dependency injection and service-container patterns, which the kernel is modelled on, An OpenAI, Azure OpenAI or other supported model endpoint to run the samples.
How long does Microsoft Semantic Kernel Documentation take?
Expect roughly Reference documentation; the quick start plus the agent and process framework guides take roughly 3-5 hours. Most learners work through it at their own pace.
What will I learn from Microsoft Semantic Kernel Documentation?
You'll learn: Build a kernel and register AI services and plugins through dependency injection; Expose existing code and APIs to a model as callable plugins and function tools; Use hooks, filters and telemetry to log, monitor and govern agent behaviour centrally; Create agents with ChatCompletionAgent, OpenAIAssistantAgent, AzureAIAgent or OpenAIResponsesAgent; Orchestrate multiple agents together using the dedicated agent orchestration package; Turn a Semantic Kernel instance into an MCP server over stdio or SSE transports; Connect vector stores and local runtimes such as Azure AI Search, Elasticsearch, Chroma, Ollama and ONNX.
Topics
Sources
This page was written from 4 sources, 2 on domains other than learn.microsoft.com.