--- title: "Microsoft Semantic Kernel Explained: Features & What Replaced It" url: "https://atlan.com/know/ai-agent/microsoft/semantic-kernel/" description: "Microsoft Semantic Kernel is an open-source SDK for LLM apps. See its components, shortcomings, and what its Agent Framework successor offers." author: "Ayswarrya G" author_role: "Contributing Writer, Data Engineering & Metadata" published: "2026-08-12" updated: "2026-08-12T00:00:00.000Z" --- --- Microsoft Semantic Kernel is an open-source, MIT-licensed SDK that lets developers combine large language models with conventional code through plugins, connectors, memory, and orchestration primitives. It ships for .NET, Python, and Java, and runs on any cloud or on-premises setup, including official Amazon Bedrock connectors, with no Azure requirement. In April 2026, the same Microsoft team shipped a production-ready successor, Agent Framework 1.0, which now carries all new feature investment while Semantic Kernel stays in maintenance mode for critical fixes and security patches. Teams already running Semantic Kernel in production don't need to rip anything out. The SDK keeps its support window, and Agent Framework was built specifically so existing plugins, connectors, and enterprise integrations carry forward rather than requiring a rewrite from scratch. | Quick facts | | |---|---| | Vendor | Microsoft | | License | MIT (open source) | | Launched | March 2023 (C#/.NET first; Python and Java followed) | | Cloud requirement | None. Runs on any cloud or on-premises; official Amazon Bedrock connectors included | | Successor | Microsoft Agent Framework 1.0, GA April 3, 2026 | | Support window | Critical bug fixes and security patches through at least April 2027 | | Core components | Kernel, Plugins, Connectors, Memory, Agents, Process Framework, Filters and telemetry | --- ## What is Microsoft Semantic Kernel? Microsoft Semantic Kernel is an open-source SDK for integrating LLMs into applications, letting developers combine AI models with conventional code through plugins, connectors, and [orchestration primitives](https://atlan.com/know/ai-agent/ai-agent-primitives/). Launched in March 2023 with C#/.NET as the primary path, it runs inside your own application on any cloud or on-premises, with Azure as its most common pairing. In April 2026, Microsoft released Agent Framework 1.0, the successor built by the same team, unifying Semantic Kernel's enterprise foundations with [AutoGen's orchestration approach](https://atlan.com/know/ai-agent/what-is-autogen/) into one SDK. **Core components of Semantic Kernel**: * **Kernel**: The central object that wires together models, plugins, and services. * **Plugins**: Native functions and prompt templates that an LLM can invoke. * **Connectors**: Adapters for models like Azure OpenAI, OpenAI, and Hugging Face. * **Memory**: [Vector store integrations](https://atlan.com/know/vector-database-vs-knowledge-graph-agent-memory/) for retrieval and semantic search. * **Agents:** Abstractions like ChatCompletionAgent for single and multi-agent chat. * **Process Framework**: Structured workflows for modeling business processes. * **Filters and telemetry**: Enterprise hooks for observability, security, and control. Semantic Kernel is an [orchestration framework](https://atlan.com/know/ai-agent/ai-agent-architecture-explained/) that solves the application side of enterprise AI. However, it doesn't resolve the missing context: what a table means, which metric definition is canonical, and which policies apply before an agent acts. Atlan's [Context Layer for AI](https://atlan.com/know/agent-context-layer/) sits beneath the orchestration layer and supplies context that agents need to deliver in production. --- ## What is Microsoft Semantic Kernel used for? Enterprise teams reach for the [open-source SDK](https://learn.microsoft.com/en-us/semantic-kernel/overview/) to give developers a structured way to embed LLM calls into real applications. Using Semantic Kernel, you could embed copilots into existing products, automate tasks, ground responses in company data via [RAG](https://atlan.com/know/ai-agent/agent-context-layer-vs-rag/), and run [multi-agent systems](https://atlan.com/know/ai-agent/debugging-multi-agent-systems/). It ships as a package for .NET, Python, and Java, and runs inside your own application on any cloud or on-premises. ### Microsoft Semantic Kernel architecture: A quick overview Its central abstraction was the [Kernel object](https://learn.microsoft.com/en-us/semantic-kernel/concepts/kernel?pivots=programming-language-csharp), a hub that wired together plugins, planners, memory, and model connectors. ![An overview of the Kernel object](/images/semantic-kernel/1-kernel-object-overview.webp){width=602 height=293} **Source**: [Microsoft Learn](https://learn.microsoft.com/en-us/semantic-kernel/concepts/kernel?pivots=programming-language-csharp) Around that core sat the plumbing .NET teams expected: dependency injection, filters, telemetry hooks, and connector support for vector stores and embedding services. Three foundational characteristics of the Semantic Kernel were: * **Plugins as the unit of capability**: Plugins, originally called skills, were collections of native functions and prompt templates that an LLM could invoke. * **Model-agnostic connectors**: One codebase could swap between Azure OpenAI, OpenAI, Ollama, and other providers. * **Enterprise plumbing by default**: Type safety, filters, and telemetry made it viable for regulated environments where LangChain-style experimentation felt risky. ### Is the Semantic Kernel still in use, or has it been replaced? Both, and the distinction matters for planning. On April 3, 2026, Microsoft [shipped Microsoft Agent Framework 1.0 (for .NET and Python)](https://devblogs.microsoft.com/agent-framework/microsoft-agent-framework-version-1-0/) as the production-ready release with stable APIs. Microsoft's own documentation calls Agent Framework "the successor to Semantic Kernel for building AI agents," and describes it as unifying Semantic Kernel's enterprise foundations with AutoGen's orchestration innovations into a single SDK. Created by the same teams, Agent Framework combines AutoGen's agent abstractions with Semantic Kernel's enterprise features, such as session-based state management, type safety, middleware, and telemetry. ![An overview of the Agent Framework](/images/semantic-kernel/2-agent-framework-overview.webp){width=602 height=308} **Source**: [Microsoft Dev Blogs](https://devblogs.microsoft.com/foundry/introducing-microsoft-agent-framework-the-open-source-engine-for-agentic-ai-apps/) [Key capabilities of the Agent Framework](https://learn.microsoft.com/en-us/agent-framework/overview/?pivots=programming-language-csharp) include: * **Agents**: [Autonomous agents](https://atlan.com/know/ai-agent/what-is-an-ai-agent/) using LLMs to process inputs and generate responses. Offers support for Microsoft Foundry, Anthropic, Azure OpenAI, OpenAI, and Ollama. * **Harness**: An agent that [can reason](https://atlan.com/know/ai-agent/what-is-an-agent-loop/), with capabilities for long, multi-step tasks. * **Workflows**: Graph-based workflows connecting agents and [functions](https://atlan.com/know/ai-agent/ai-agent-tool-use/) for multi-step tasks. The timeline in brief: 1. **October 2025:** Microsoft announced the public preview of Agent Framework, published as agent-framework on PyPI and Microsoft.Agents.AI on NuGet. 2. **February 19, 2026:** Agent Framework reached Release Candidate for .NET and Python, with a stable API surface and all v1.0 features complete. 3. **April 3, 2026:** Version 1.0 went GA with stable APIs, a long-term support commitment, and native MCP and A2A interoperability. Semantic Kernel itself [hasn't vanished, and a substantial number of developers](https://devblogs.microsoft.com/agent-framework/semantic-kernel-and-microsoft-agent-framework/) still use it. For new deployments, the Semantic Kernel GitHub repository now points developers to Agent Framework as the [enterprise-ready successor](https://atlan.com/know/ai-agent/ai-agent-scaling-in-production/). The CIO's Guide to Context Graphs See how context graphs connect the definitions, lineage, and policy an agent needs across your orchestration stack. Get the CIO Context Guide --- ## What are the main features of Semantic Kernel? The [top features of Semantic Kernel](https://learn.microsoft.com/en-us/semantic-kernel/concepts/semantic-kernel-components?pivots=programming-language-csharp), most of which carry forward into Agent Framework, include: * **Functions and plugins**: Named containers of functions the model can invoke, created from native code, OpenAPI specs, text search implementations for RAG, or prompt templates. * **AI service connectors**: An [abstraction layer](https://atlan.com/know/ai-agent/ai-platform-architecture/) across Azure OpenAI, OpenAI, and Amazon Bedrock, spanning chat, text generation, embeddings, image, and audio services. * **​​Vector Store (memory) connectors**: A common interface over vector stores like Azure AI Search, Elasticsearch, and other stores, with vector search exposable as a plugin for retrieval. * **Multi-provider connectors:** Built-in support for OpenAI, Azure OpenAI, NVIDIA, and more. * **Prompt templates**: Templates that combine AI instructions, user input, and function output, usable as the start of a chat flow or as [callable plugin functions](https://atlan.com/know/ai-agent/ai-agent-skills/what-are-agent-skills/). * **Filters**: Interception points before and after function invocation and prompt rendering, which is where teams implement [security controls](https://atlan.com/know/ai-agent-access-control/) and [responsible AI checks](https://atlan.com/know/ai-agent-governance/). * **[Observability and telemetry](https://atlan.com/know/ai-agent-observability/)**: Logs, metrics, and traces emitted through OpenTelemetry, so agent behavior is inspectable with standard monitoring tools. * **Agent and process frameworks:** [Multi-agent orchestration](https://atlan.com/know/ai-agent/ai-agent-context/context-management-multi-agent-systems/) with collaborating specialist agents, including a Process Framework for structured business workflows. * **MCP support**: Later releases added [Model Context Protocol](https://atlan.com/know/what-is-model-context-protocol/) integration, letting kernel-based agents consume tools and context from external servers. Every one of these features still assumes the context an agent needs already exists somewhere accessible. None of them generate it, which is the gap the next section gets into. --- ## What are the shortcomings of Semantic Kernel? Some of Semantic Kernel's shortcomings disappear with Agent Framework, but some follow you across the migration. The biggest shortcomings of Semantic Kernel include: * **Feature development has ended**: Microsoft addresses critical bugs and security issues in Semantic Kernel, but new features land only in Agent Framework. * **The planner detour**: Prompt-based planners were retired as native function calling proved more reliable. * **Fragmentation with AutoGen**: Developers were confused about which framework to use, [features overlapped](https://atlan.com/know/ai-agent/agent-sprawl/), and the community was split. The launch of AgentFramework resolved this directly. * **Verbose multi-agent orchestration**: In Semantic Kernel, you write your own logic for which agent speaks next, when the conversation ends, and how information passes between agents. Agent Framework ships these patterns (sequential, concurrent, group chat, handoff, and [magentic](https://learn.microsoft.com/en-us/semantic-kernel/frameworks/agent/agent-orchestration/magentic?pivots=programming-language-csharp)) ready-made * **API churn across versions**: The road from experimental skills to plugins to agents involved breaking changes that made long-lived codebases expensive to maintain. Microsoft's [migration guide](https://learn.microsoft.com/en-us/agent-framework/migration-guide/from-semantic-kernel/) reduces the cost, but doesn't remove it. * **Context remains unsolved by design**: Agent Framework gives agents better sockets for context, including MCP integrations and context providers. However, it still doesn't provide the governed context itself: [whether a metric definition is current](https://atlan.com/know/context-drift-ai-agents/), which table is the trusted source, and what policies apply before an agent acts. That last gap is the one no framework version will ever close, because it sits outside the orchestration layer entirely. A perfectly orchestrated agent reasoning over a stale metric definition produces a confident wrong answer, faster and at greater scale than before. Atlan's [enterprise context layer for AI](https://atlan.com/know/why-ai-agents-need-an-enterprise-context-layer/) is built to close that gap. Context Gap Calculator Score how much of your agent stack's context gap is architecture versus tooling before you plan the next migration. Try the Context Gap Calculator --- ## How does Atlan fill the gap for Semantic Kernel (and Agent Framework) users? Atlan is the Context Layer for AI. Orchestration frameworks like Semantic Kernel and Agent Framework decide how agents plan, call tools, and coordinate. Atlan supplies what those agents need to be right: governed definitions, relationships, lineage, quality signals, and policy rules, served at inference time. Key capabilities include: * **Context Lakehouse**: Atlan stores enterprise context in an Iceberg-native knowledge architecture with vector-native search, so agents retrieve meaning rather than keywords. It's queryable over standard SQL, MCP, and APIs. * **MCP server**: Atlan's MCP server [plugs into your existing stack](https://atlan.com/know/ai-agent/how-to-give-ai-agents-access-to-enterprise-data/) and lets agents check what an asset means, whether it is trusted, and which policies apply in a single call before anything executes. * **Context Engineering Studio**: Data teams build, test, and version the context their agents consume. Existing dashboards and production queries become an eval suite, so [context ships with tests the same way code does](https://atlan.com/know/ai-agent/context-versioning-for-ai-agents/). * **Context Agents**: AI agents that generate and enrich descriptions, classifications, and quality rules continuously, so the [context layer](https://atlan.com/know/what-is-a-context-graph/) stays current as the data estate changes underneath it. The practical payoff is framework independence. Teams migrating from Semantic Kernel to Agent Framework rewrite orchestration code, but context built in Atlan [moves with them unchanged](https://atlan.com/know/ai-agent/context-portability/), because it lives behind a vendor-agnostic open protocol. See Context Layer Demos Live Watch how Atlan serves governed context to agents at inference time, independent of which orchestration framework sits on top. Watch the Live Demo Series --- ## Real stories from real customers: context beyond the orchestration layer Neither of these teams runs on Semantic Kernel specifically, but both hit the same wall every enterprise agent stack eventually hits: the framework can call a tool, but it can't tell you whether the data behind that tool is trustworthy. That's the layer they built with Atlan instead of inside their orchestration code. "Atlan captures Workday's shared language to be leveraged by AI via its MCP server. As part of Atlan's AI labs, we're co-building the semantic layer that AI needs." Joe DosSantos, VP Enterprise Data & Analytics, Workday Watch Now "Atlan is much more than a catalog of catalogs. It's more of a context operating system…Atlan enabled us to easily activate metadata for everything from discovery in the marketplace to AI governance to data quality to an MCP server delivering context to AI models." Sridher Arumugham, Chief Data & Analytics Officer, DigiKey Watch Now --- ## Moving forward with Microsoft Semantic Kernel Semantic Kernel brought LLMs into .NET shops with the engineering discipline those environments demand. Its plugin model shaped how teams think about AI capabilities, and it did so years before most competing frameworks had settled on a stable API. It has now paved the way for Agent Framework 1.0, which has Semantic Kernel as its foundation layer and AutoGen-style orchestration running on top. Existing systems in Semantic Kernel have a support window through at least April 2027, but new builds should start on Agent Framework, since that is where Microsoft's engineering investment, documentation, and long-term support commitment now sit. For data teams though, both frameworks still don't plug the context gap. Their biggest concern remains unsolved: where the governed context will live, who keeps it current, and how it gets tested before an agent relies on it. Context must be seen as a [living infrastructure](https://atlan.com/know/context-engineering-for-ai-agents/) that's versioned, tested, and served over open protocols, not a folder of prompt strings someone updates when they remember to. That is the foundation Atlan's enterprise context layer for AI provides, regardless of whichever Microsoft framework sits on top, and regardless of whether the next migration happens in six months or six years. [Book a demo](https://atlan.com/forms/talk-to-sales-contact/) to see it in action. --- ## FAQs about Microsoft Semantic Kernel ### 1. What is Microsoft Semantic Kernel used for? Semantic Kernel is used to integrate large language models into applications. Developers combine AI models with conventional code through plugins, connect to providers like Azure OpenAI, add retrieval through vector stores, and orchestrate multi-step AI workflows, all with enterprise features like telemetry and type safety. ### 2. Is Semantic Kernel deprecated? Not formally deprecated, but superseded. Microsoft continues to support it with critical bug fixes and security patches for at least one year after Agent Framework reached general availability, which puts committed support through at least April 2027. Critical bug fixes and security patches continue, while new feature development happens in Agent Framework. ### 3. What replaced Semantic Kernel? Microsoft Agent Framework 1.0, released April 3, 2026, unifies Semantic Kernel and AutoGen into a single production SDK. Semantic Kernel serves as the foundation layer, with graph-based multi-agent orchestration built on top. ### 4. What is the difference between Semantic Kernel and AutoGen? Semantic Kernel was the .NET-first framework focused on plugins, connectors, and enterprise integration. AutoGen was the Python-friendly framework for multi-agent conversation patterns. The two teams shipped one framework, Microsoft Agent Framework, as Semantic Kernel's successor, unifying it with AutoGen's orchestration model. ### 5. Does Semantic Kernel support MCP? Yes. Semantic Kernel supports extending agents with Model Context Protocol servers alongside native code functions, prompt templates, and OpenAPI specs. This lets agents consume tools and context from external systems, including governed context layers, through an open standard. ### 6. Is Microsoft Agent Framework still in preview? No, it is generally available. Agent Framework entered public preview in October 2025, reached Release Candidate on February 19, 2026, and shipped version 1.0 on April 3, 2026 with stable APIs and a long-term support commitment from Microsoft. ### 7. Should I migrate from Semantic Kernel to Agent Framework? For new projects, yes, start on Agent Framework directly, since it is now GA and receives all new investment. For production systems, plan migration within the support window using Microsoft's official migration guides. Stable systems that don't need new orchestration patterns can reasonably stay on Semantic Kernel for now. --- ## Sources 1. Introduction to Semantic Kernel. Microsoft Learn. https://learn.microsoft.com/en-us/semantic-kernel/overview/ 2. Microsoft Agent Framework overview. Microsoft Learn. https://learn.microsoft.com/en-us/agent-framework/overview/ 3. Semantic Kernel and Microsoft Agent Framework. Microsoft Agent Framework Blog. https://devblogs.microsoft.com/agent-framework/semantic-kernel-and-microsoft-agent-framework/ 4. Semantic Kernel components. Microsoft Learn. https://learn.microsoft.com/en-us/semantic-kernel/concepts/semantic-kernel-components 5. Migrate your Semantic Kernel and AutoGen projects to Microsoft Agent Framework Release Candidate. Microsoft Agent Framework Blog. https://devblogs.microsoft.com/agent-framework/migrate-your-semantic-kernel-and-autogen-projects-to-microsoft-agent-framework-release-candidate/ 6. Semantic Kernel to Microsoft Agent Framework migration guide. Microsoft Learn. https://learn.microsoft.com/en-us/agent-framework/migration-guide/from-semantic-kernel/ 7. microsoft/semantic-kernel. GitHub. https://github.com/microsoft/semantic-kernel