Semantic Kernel

Semantic Kernel is Microsoft's SDK for building LLM-powered applications with enterprise patterns — plugin system for tool definition, multiple planner strategies (sequential, step-wise), memory connectors for RAG, and agent patterns. Strong fit for enterprise teams on .NET/C# stacks or Azure AI services. Increasingly Python-capable (SK 1.0+). The plugin model is particularly clean for defining structured LLM tools. Confidence is docs-derived.

.NET: dotnet add package Microsoft.SemanticKernel. Python: pip install semantic-kernel. Kernel = the central orchestration object. Plugins: @kernel_function decorated Python functions or C# classes. Planner: selects and sequences plugins to achieve goals. Memory: semantic memory via embedding stores (Azure AI Search, Chroma, Pinecone). Agent: KernelAgent with tool-calling loop. Azure OpenAI and OpenAI both supported as LLM backends.

No SK-specific API key — open-source library. LLM provider keys configured via Kernel builder (AzureOpenAI, OpenAI, HuggingFace). Azure integration uses managed identity or Azure credentials. Open-source MIT license. Memory store credentials managed per-connector.

.NET exception handling for LLM and plugin failures. Python: exception-based error propagation. Planner failures: PlanningException with details. Memory store connectivity failures surface at kernel invocation time. Active Microsoft engineering team maintaining all three SDK variants. Stable v1.0 API in .NET and Python.

learn.microsoft.com/semantic-kernel provides comprehensive Microsoft Learn documentation — quickstarts for .NET and Python, plugin authoring, planner guides, memory connectors, and agent patterns. Getting started: straightforward for both .NET and Python. Strong Azure AI integration documentation. Community via SK Discord and GitHub (Microsoft/semantic-kernel). Regular releases.