Semantic Kernel

1.39.4 · active · verified Sat Feb 28

Microsoft's open-source SDK for integrating LLMs into applications. Model-agnostic (OpenAI, Azure OpenAI, Google, Hugging Face, Mistral, etc.), supports Python, C#, and Java. Provides plugins (tools), prompt templating, multi-agent orchestration (AgentChat framework), and vector store integrations for RAG. Production-stable as of v1.x. Currently the primary Microsoft-maintained AI framework (AutoGen is maintenance-only; Microsoft Agent Framework is the future successor).

Warnings

breaking Azure Identity fallback authentication removed. In older versions, AzureChatCompletion would silently fall back to azure-identity DefaultAzureCredential if no api_key was provided. This implicit fallback has been removed. Code relying on it will now raise errors on import or at runtime.
Fix: Install azure-identity explicitly (pip install azure-identity or pip install semantic-kernel[azure]) and pass credentials explicitly: AzureChatCompletion(ad_token_provider=...) or AzureChatCompletion(api_key=...)
breaking SemanticTextMemory and MemoryStoreBase are deprecated and will be removed. The entire old memory store architecture (semantic_kernel.memory, semantic_kernel.connectors.memory_stores) has been superseded by the new VectorStore API.
Fix: Migrate to semantic_kernel.data.vector: use @vectorstoremodel, VectorStoreField, and connector-specific VectorStore classes (e.g., from semantic_kernel.connectors.memory import ChromaCollection). See: https://learn.microsoft.com/en-us/semantic-kernel/support/migration/vectorstore-python-june-2025
breaking Google connector migrated from google-generativeai SDK to the new google-genai SDK (v1.39.0, Feb 2026). If using the Google connector, the underlying dependency has changed.
Fix: Upgrade to 1.39.0+. Uninstall google-generativeai if no longer needed. Install semantic-kernel[google] which now pulls in google-genai.
breaking Template argument encoding now throws on complex types. Previously, complex types (custom classes, anonymous types, collections) passed as prompt template arguments were silently not encoded. Now an exception is raised if automatic encoding is enabled (the default) and a complex type is passed.
Fix: Serialize complex types to strings before passing as template arguments, or disable automatic encoding.
breaking sk_function decorator removed. The old @sk_function decorator from pre-1.0 is gone. All plugin methods must use @kernel_function.
Fix: Replace @sk_function with @kernel_function from semantic_kernel.functions.
breaking SKContext removed. Replaced by KernelArguments. Any code importing or using SKContext will fail.
Fix: Replace SKContext with KernelArguments from semantic_kernel.functions.
gotcha azure-identity is NOT included in the base pip install semantic-kernel. Importing AzureChatCompletion (which requires Entra ID support) will fail with ImportError if azure-identity is not installed separately.
Fix: pip install semantic-kernel[azure] or pip install azure-identity
gotcha Microsoft Agent Framework (successor to both AutoGen and Semantic Kernel) has reached Release Candidate status (Feb 2026). Semantic Kernel remains actively maintained but Microsoft is directing new agentic projects toward Agent Framework.
Fix: For new agentic projects, evaluate Microsoft Agent Framework. Existing SK code is stable and supported. Migration guide: https://aka.ms/autogen-to-af
gotcha Pydantic v1 is not supported. semantic-kernel requires Pydantic v2. If your project pins Pydantic v1 (e.g., via an older FastAPI or LangChain version), there will be a dependency conflict.
Fix: Upgrade to Pydantic v2. Check for other dependencies pinning Pydantic v1.

Install

pip install --upgrade semantic-kernel Minimal install (OpenAI/Azure OpenAI)
pip install --upgrade semantic-kernel[azure] With Azure extras (azure-identity, azure-ai-inference, azure-search-documents)
pip install --upgrade semantic-kernel[google] With Google Gemini connector
pip install --upgrade semantic-kernel[all] All connectors and extras

Imports

Kernel
```
from semantic_kernel import Kernel
```
Import directly from the package root.
AzureChatCompletion
```
from semantic_kernel.connectors.ai.open_ai import AzureChatCompletion
```
Azure OpenAI connector lives under the openai module, not a separate azure module.
kernel_function decorator
```
from semantic_kernel.functions import kernel_function
```
sk_function was the old decorator name (pre-1.0). Removed. Use @kernel_function.
SemanticTextMemory (DEPRECATED)
```
from semantic_kernel.data.vector import VectorStoreField, vectorstoremodel
```
SemanticTextMemory and MemoryStoreBase are deprecated. Migrate to VectorStore/VectorStoreCollection.
KernelArguments
```
from semantic_kernel.functions import KernelArguments
```
SKContext was removed in 1.0. Use KernelArguments.

Quickstart

Basic chat completion with OpenAI

import asyncio
from semantic_kernel import Kernel
from semantic_kernel.connectors.ai.open_ai import OpenAIChatCompletion
from semantic_kernel.connectors.ai.open_ai import OpenAIChatPromptExecutionSettings
from semantic_kernel.contents import ChatHistory

async def main():
    kernel = Kernel()
    kernel.add_service(OpenAIChatCompletion(ai_model_id="gpt-4o"))

    history = ChatHistory(system_message="You are a helpful assistant.")
    history.add_user_message("Hello, what can you do?")

    service = kernel.get_service(type=OpenAIChatCompletion)
    settings = OpenAIChatPromptExecutionSettings()
    response = await service.get_chat_message_content(chat_history=history, settings=settings, kernel=kernel)
    print(response.content)

asyncio.run(main())

view raw JSON →