Microsoft's open-source SDK for building AI agents that combine LLMs with semantic plugins and memory.
Semantic Kernel is an open-source SDK developed by Microsoft for building AI applications and agents that orchestrate large language models with plugins, memory, and planning capabilities. It provides a framework for defining 'semantic functions' — LLM prompts that can be composed and chained — alongside 'native functions' — traditional code — enabling developers to build complex AI workflows that combine natural language understanding with deterministic business logic.
Semantic Kernel has become one of the most widely adopted frameworks for enterprise AI agent development. Its integration with Azure OpenAI, its plugin architecture (which is semantically compatible with MCP), and its memory management capabilities make it a natural choice for organizations building AI agents on Microsoft infrastructure. In 2026, Semantic Kernel's semantic memory features — storing and retrieving information based on semantic similarity — are widely used for enterprise knowledge management.
Semantic Kernel organizes AI capabilities into Skills (collections of functions), Plugins (external integrations), Memory (semantic storage and retrieval), and Planners (AI-generated execution plans). The kernel orchestrates these components, routing requests to the appropriate functions and managing context. Semantic memory uses vector embeddings to store and retrieve information by meaning, enabling AI agents to remember and recall relevant context from past interactions.
An enterprise builds a procurement AI agent using Semantic Kernel. The agent has plugins for querying the ERP system, searching the vendor knowledge base, and sending approval requests. When asked to 'Find the best vendor for 10,000 units of component X within our approved vendor list, considering our Q2 budget constraints,' the Semantic Kernel planner automatically sequences the required plugin calls and synthesizes the result.