Semantic Kernel is already a migration guide to its own replacement

What it does

Semantic Kernel is a model-agnostic SDK for building, orchestrating, and deploying AI agents and multi-agent systems. It abstracts connections to various LLMs—OpenAI, Azure OpenAI, Hugging Face, and others—and wraps them in an agent framework with plugins, memory, and planning. It also bundles integrations for vector databases like Azure AI Search and Elasticsearch, plus a Process Framework for structured business workflows.

The interesting bit

The README’s most prominent feature is the announcement that Semantic Kernel is now the Microsoft Agent Framework, complete with a migration guide. That makes this repository less a destination and more a well-documented waypoint; the code still works, but the long-term support and stable APIs have officially moved to the successor project. It is rare to see an open-source project with 28,000 stars so candidly announce its own obsolescence on its own front page.

Key highlights

• Supports Python, .NET, and Java, with local deployment via Ollama, LMStudio, and ONNX.
• Extends agents through native code functions, OpenAPI specs, or Model Context Protocol (MCP).
• Integrates with vector databases including Azure AI Search, Elasticsearch, and Chroma.
• Handles multimodal inputs: text, vision, and audio.
• Includes a Process Framework for modeling complex business workflows.

Caveats

• The README explicitly positions Semantic Kernel as the predecessor to the Microsoft Agent Framework, which is now the home of stable APIs and long-term support.
• The front page is dominated by migration guidance rather than new feature announcements.

Verdict

Developers already using Semantic Kernel should treat this repo as a transition zone and follow the migration guide. If you are starting a new project, the Microsoft Agent Framework is likely the safer long-term bet.