Important
Semantic Kernel is now Microsoft Agent Framework! Microsoft Agent Framework (MAF) is the enterprise‑ready successor to Semantic Kernel. Microsoft Agent Framework is now available at version 1.0 as a production-ready release: stable APIs, and a commitment to long-term support. Whether you're building a single assistant or orchestrating a fleet of specialized agents, Microsoft Agent Framework 1.0 gives you enterprise-grade multi-agent orchestration, multi-provider model support, and cross-runtime interoperability via A2A and MCP.
Learn more about Semantic Kernel and Agent Framework here: Semantic Kernel and Microsoft Agent Framework on the Agent Framework blog, and try out the Semantic Kernel migration guide.
Highlights
- Flexible Agent Framework: build, orchestrate, and deploy AI agents and multi-agent systems
- Multi-Agent Systems: Model workflows and collaboration between AI specialists
- Plugin Ecosystem: Extend with Python, OpenAPI, Model Context Protocol (MCP), and more
- LLM Support: OpenAI, Azure OpenAI, Hugging Face, Mistral, Google AI, ONNX, Ollama, NVIDIA NIM, and others
- Vector DB Support: Azure AI Search, Elasticsearch, Chroma, and more
- Process Framework: Build structured business processes with workflow modeling
- Multimodal: Text, vision, audio
pip install --upgrade semantic-kernel
# Optional: Add integrations
pip install --upgrade semantic-kernel[hugging_face]
pip install --upgrade semantic-kernel[all]Supported Platforms:
- Python: 3.10+
- OS: Windows, macOS, Linux
Set as environment variables, or create a .env file at your project root:
OPENAI_API_KEY=sk-...
OPENAI_CHAT_MODEL_ID=...
...
AZURE_OPENAI_API_KEY=...
AZURE_OPENAI_ENDPOINT=...
AZURE_OPENAI_CHAT_DEPLOYMENT_NAME=...
...You can also override environment variables by explicitly passing configuration parameters to the AI service constructor:
chat_service = AzureChatCompletion(
api_key=...,
endpoint=...,
deployment_name=...,
api_version=...,
)See the following setup guide for more information.
Create prompt functions and invoke them via the Kernel:
import asyncio
from semantic_kernel import Kernel
from semantic_kernel.connectors.ai.open_ai import OpenAIChatCompletion
from semantic_kernel.functions import KernelArguments
kernel = Kernel()
kernel.add_service(OpenAIChatCompletion())
prompt = """
1) A robot may not injure a human being...
2) A robot must obey orders given it by human beings...
3) A robot must protect its own existence...
Give me the TLDR in exactly {{$num_words}} words."""
async def main():
result = await kernel.invoke_prompt(prompt, arguments=KernelArguments(num_words=5))
print(result)
asyncio.run(main())
# Output: Protect humans, obey, self-preserve, prioritized.You can use the AI service classes directly for advanced workflows:
import asyncio
import asyncio
from semantic_kernel.connectors.ai.open_ai import OpenAIChatCompletion, OpenAIChatPromptExecutionSettings
from semantic_kernel.contents import ChatHistory
async def main():
service = OpenAIChatCompletion()
settings = OpenAIChatPromptExecutionSettings()
chat_history = ChatHistory(system_message="You are a helpful assistant.")
chat_history.add_user_message("Write a haiku about Semantic Kernel.")
response = await service.get_chat_message_content(chat_history=chat_history, settings=settings)
print(response.content)
"""
Output:
Thoughts weave through context,
Semantic threads interlace—
Kernel sparks meaning.
"""
asyncio.run(main())Add Python functions as plugins or Pydantic models as structured outputs;
Enhance your agent with custom tools (plugins) and structured output:
import asyncio
from typing import Annotated
from pydantic import BaseModel
from semantic_kernel.agents import ChatCompletionAgent
from semantic_kernel.connectors.ai.open_ai import AzureChatCompletion, OpenAIChatPromptExecutionSettings
from semantic_kernel.functions import kernel_function, KernelArguments
class MenuPlugin:
@kernel_function(description="Provides a list of specials from the menu.")
def get_specials(self) -> Annotated[str, "Returns the specials from the menu."]:
return """
Special Soup: Clam Chowder
Special Salad: Cobb Salad
Special Drink: Chai Tea
"""
@kernel_function(description="Provides the price of the requested menu item.")
def get_item_price(
self, menu_item: Annotated[str, "The name of the menu item."]
) -> Annotated[str, "Returns the price of the menu item."]:
return "$9.99"
class MenuItem(BaseModel):
# Used for structured outputs
price: float
name: str
async def main():
# Configure structured outputs format
settings = OpenAIChatPromptExecutionSettings()
settings.response_format = MenuItem
# Create agent with plugin and settings
agent = ChatCompletionAgent(
service=AzureChatCompletion(),
name="SK-Assistant",
instructions="You are a helpful assistant.",
plugins=[MenuPlugin()],
arguments=KernelArguments(settings),
)
response = await agent.get_response("What is the price of the soup special?")
print(response.content)
# Output:
# The price of the Clam Chowder, which is the soup special, is $9.99.
asyncio.run(main())You can explore additional getting started agent samples here.
Treat model-generated function calls and arguments as untrusted input. Prompts, function descriptions, and a model's decision to call a tool are not authorization controls. The application must authorize the caller, target resource, and arguments before performing a sensitive operation.
FunctionChoiceBehavior.Auto() without filters makes all functions registered in the kernel available
for automatic invocation. Register only the functions needed for the task, or use an explicit, nonempty
allowlist:
from semantic_kernel.connectors.ai import FunctionChoiceBehavior
from semantic_kernel.connectors.ai.open_ai import OpenAIChatPromptExecutionSettings
settings = OpenAIChatPromptExecutionSettings(
function_choice_behavior=FunctionChoiceBehavior.Auto(
filters={"included_functions": ["MenuPlugin-get_specials"]}
)
)Function names in included_functions use the PluginName-function_name format. Use the plugin name
and function name actually registered in your kernel. With filters configured, the chat completion
automatic invocation path checks the requested function against the filtered kernel metadata before
invoking it. Keep the kernel's registered functions and the behavior's filters unchanged during a request;
this check is not a snapshot of the tools sent to the model. An empty inclusion list is not a deny-all
policy: empty lists do not restrict function selection.
Function-choice filters restrict which functions can be selected; they do not authorize access to
data or validate whether an operation is permitted for a particular user. Enforce those checks in the
function or in a function invocation filter.
An automatic function invocation filter
can also inspect a proposed call and decline to invoke it by not calling next(context). Setting
context.terminate after await next(context) stops subsequent processing; it does not undo the call.
For application approval before execution, use FunctionChoiceBehavior.Auto(auto_invoke=False, filters=...),
then inspect and authorize each returned call before invoking it. If you manually call
Kernel.invoke_function_call, pass the same FunctionChoiceBehavior as its function_behavior argument
to retain function-name filtering. Omitting that argument skips that filtering, and passing
execution_settings alone does not enable it. Direct calls such as Kernel.invoke also require the
application's own authorization; function-choice filters are not a kernel-wide access-control policy.
Coordinate a group of agents to iteratively solve a problem or refine content together:
import asyncio
from semantic_kernel.agents import ChatCompletionAgent, GroupChatOrchestration, RoundRobinGroupChatManager
from semantic_kernel.agents.runtime import InProcessRuntime
from semantic_kernel.connectors.ai.open_ai import AzureChatCompletion
def get_agents():
return [
ChatCompletionAgent(
name="Writer",
instructions="You are a creative content writer. Generate and refine slogans based on feedback.",
service=AzureChatCompletion(),
),
ChatCompletionAgent(
name="Reviewer",
instructions="You are a critical reviewer. Provide detailed feedback on proposed slogans.",
service=AzureChatCompletion(),
),
]
async def main():
agents = get_agents()
group_chat = GroupChatOrchestration(
members=agents,
manager=RoundRobinGroupChatManager(max_rounds=5),
)
runtime = InProcessRuntime()
runtime.start()
result = await group_chat.invoke(
task="Create a slogan for a new electric SUV that is affordable and fun to drive.",
runtime=runtime,
)
value = await result.get()
print(f"Final Slogan: {value}")
# Example Output:
# Final Slogan: "Feel the Charge: Adventure Meets Affordability in Your New Electric SUV!"
await runtime.stop_when_idle()
if __name__ == "__main__":
asyncio.run(main())For orchestration-focused examples, see these orchestration samples.
- Getting Started with Agents: Practical agent orchestration and tool use
- Getting Started with Processes: Modeling structured workflows with the Process framework
- Concept Samples: Advanced scenarios, integrations, and SK patterns
- Getting Started Notebooks: Interactive Python notebooks for rapid experimentation