What it does
Semantic Kernel is a software development kit from Microsoft for creating AI agents and systems of cooperating agents. It is not tied to one model vendor: the repository lists built-in support for OpenAI, Azure OpenAI, Hugging Face and NVIDIA, and says local models can run through Ollama, LMStudio or ONNX. Agents can use plugins built from native code functions, prompt templates, OpenAPI specs or Model Context Protocol servers, and can draw on memory and planning. The README also covers vector database integrations such as Azure AI Search, Elasticsearch and Chroma, text, vision and audio inputs, and a process framework for structured business workflows. Install it with pip for Python or dotnet add package for .NET; Java has its own build instructions. The README lists Python 3.10+, .NET 10.0+ and JDK 17+, and support for Windows, macOS and Linux. A notice at the top of the page says Semantic Kernel is now Microsoft Agent Framework, describes that framework as its production-ready successor and points to a migration guide.
Capabilities
- SDK for Python, .NET and Java
- Connectors for OpenAI, Azure OpenAI, Hugging Face, NVIDIA and local runtimes such as Ollama
- Agents with plugins, memory and planning
- Multi-agent orchestration with specialist agents
- Plugins from native code, prompt templates, OpenAPI specs or MCP
- Vector database integrations and a process framework for workflows
When to use it
- Building agents in a C#, Python or Java codebase that should work with more than one model provider
- Giving an agent tools through plugins, including OpenAPI or MCP based ones
- Maintaining an existing Semantic Kernel application or planning a move to Microsoft Agent Framework with the migration guide
Alternatives to explore
- Microsoft Agent Framework
