Semantic Kernel is Microsoft's open-source SDK for building AI agents and plugins across C#, Python, and Java, now succeeded by Microsoft Agent Framework.
Semantic Kernel is Microsoft’s open-source SDK for connecting large language models to your existing code in C#, Python, or Java. It works as middleware: you describe your functions as plugins, and the model decides when to call them, so you get AI features without rewriting your application logic. Microsoft has since positioned Semantic Kernel’s successor, Microsoft Agent Framework, as the enterprise-ready path forward, but Semantic Kernel remains actively maintained and widely deployed. For AI developers building agents inside a Microsoft or .NET-heavy stack, it is still one of the fastest ways to add tool-calling, memory, and multi-agent orchestration to production software without switching ecosystems.
Semantic Kernel centers on a few core concepts that stay consistent across all three supported languages.
Kernel: The central object that registers your AI services and plugins, then routes requests between your code and the model.
Plugins: Existing functions, whether native code, prompt templates, or OpenAPI specs, that you expose to the model as callable tools.
Function calling: When a user sends a request, the model decides which plugin function to call, Semantic Kernel executes it, and the result gets passed back to the model to finish the response.
Agents: A kernel, a set of instructions, and a set of plugins wrapped into a reusable agent object that can hold a conversation and take multi-step actions.
Multi-agent orchestration: Several specialist agents, for example a billing agent and a refund agent, sit behind a triage agent that routes each incoming request to the right specialist.
Connectors: Swap the underlying model, such as OpenAI, Azure OpenAI, Hugging Face, or a local model through Ollama, without changing your application code.
Customer support triage system: Route incoming tickets to the right specialist agent, such as billing, refunds, or technical support, with each agent handling its own domain logic independently.
Internal knowledge assistant: Expose company APIs and documents as plugins so employees can ask questions in plain language and trigger actions like creating a ticket or checking an order status.
DevOps helper: Build an agent that reads application logs, explains errors in plain English, and suggests or triggers remediation steps through connected tools.
Sales research copilot: Combine a web search plugin with a CRM plugin so the agent can research a prospect and log its findings directly into your pipeline.
Structured data extraction pipeline: Use function calling with typed outputs to pull consistent, validated data out of unstructured text at scale.
Cross-platform enterprise app: Ship the same agent logic in a C# backend and a Python data pipeline, since Semantic Kernel keeps its core concepts consistent across both languages.
Semantic Kernel is used to build AI agents that call your existing code, APIs, and data through plugins. Common uses include customer support triage systems, internal knowledge assistants, DevOps helpers that explain log errors, and multi-agent workflows where specialist agents collaborate on a single business process.
Yes. Semantic Kernel is open-source under the MIT license, so you can install it with pip, NuGet, or Maven and run it in your own environment at no cost. You still pay separately for whatever AI model or cloud service you connect it to, such as Azure OpenAI or OpenAI's API.
Microsoft Agent Framework is the enterprise-ready successor to Semantic Kernel, combining its orchestration model with AutoGen's multi-agent capabilities. Semantic Kernel remains actively maintained and widely used in production, but Microsoft now points new enterprise projects toward Agent Framework and provides a migration guide for teams that want to move over.
Model Context Protocol (MCP) is an open standard from Anthropic for connecting AI models to external tools, data sources, and services through a uniform interface that works across any compliant LLM client.
Framework
Haystack is deepset's open-source framework for building production-ready RAG pipelines, AI agents, and semantic search systems in Python.
Framework
Pydantic AI is a Python-native agent framework built by the Pydantic team for building type-safe, production-ready LLM applications and agents.
Framework
DSPy is Stanford's open-source Python framework that replaces brittle prompt engineering with compiled, optimised LLM programs evaluated against measurable objectives.
FrameworkUpdates from the AI world — what shipped, what we’re using in production, and what’s worth your attention. Two emails a month, no spam.
