Capability · Framework — orchestration

Semantic Kernel

Semantic Kernel (SK) is Microsoft's production-grade orchestration SDK for building AI agents and copilots. It ships a plugin model (functions + native code), a planner, vector memory connectors, and filters for responsible-AI guardrails — all designed around .NET/C# idioms while supporting Python and Java equally. It's the default orchestration layer inside Microsoft Copilot Studio and most Azure-native agent stacks.

Framework facts

Category: orchestration
Language: C# / Python / Java
License: MIT
Repository: https://github.com/microsoft/semantic-kernel

Install

pip install semantic-kernel
# or .NET
dotnet add package Microsoft.SemanticKernel

Quickstart

import asyncio
from semantic_kernel import Kernel
from semantic_kernel.connectors.ai.open_ai import AzureChatCompletion

kernel = Kernel()
kernel.add_service(AzureChatCompletion(deployment_name='gpt-4o'))

async def main():
    result = await kernel.invoke_prompt('Summarise: {{$input}}', input='Semantic Kernel orchestrates LLMs.')
    print(result)

asyncio.run(main())

Alternatives

LangChain — Python/TS-first ecosystem
Haystack — pipeline-centric, deepset-backed
AutoGen — Microsoft's multi-agent research framework

Frequently asked questions

Semantic Kernel or AutoGen?

Both are Microsoft projects. Semantic Kernel is the production SDK — ship single-agent or small-team flows with enterprise support. AutoGen is the research playground for multi-agent patterns; many AutoGen ideas migrate into SK's Agents Framework over time.

Does Semantic Kernel lock me into Azure?

No. It has connectors for OpenAI, Anthropic, Google, Mistral, local models via Ollama, and most vector DBs. Azure integration is first-class but optional.

Sources

Semantic Kernel docs — accessed 2026-04-20
Semantic Kernel GitHub — accessed 2026-04-20