microsoft-agent-framework

pip install semantic-kernel
# Agents are included in the core package

dotnet add package Microsoft.SemanticKernel
dotnet add package Microsoft.SemanticKernel.Agents.Core
dotnet add package Microsoft.SemanticKernel.Agents.OpenAI  # For OpenAI Assistant agent

from semantic_kernel.agents import ChatCompletionAgent
from semantic_kernel import Kernel

kernel = Kernel()
# Add AI service to kernel (OpenAI, Azure OpenAI, etc.)

researcher = ChatCompletionAgent(
    kernel=kernel,
    name="Researcher",
    instructions="""You are a research analyst. Given a topic,
    find key facts, statistics, and recent developments.
    Always cite sources. Output structured findings.""",
)

writer = ChatCompletionAgent(
    kernel=kernel,
    name="Writer",
    instructions="""You are a technical writer. Take research findings
    and produce clear, engaging content. Use simple language.
    Target audience: software developers.""",
)

from semantic_kernel.functions import kernel_function

class WebSearch:
    @kernel_function(description="Search the web for information")
    def search(self, query: str) -> str:
        # Implement search logic
        return f"Results for: {query}"

class Calculator:
    @kernel_function(description="Perform mathematical calculations")
    def calculate(self, expression: str) -> str:
        return str(eval(expression))  # Use safe eval in production

kernel.add_plugin(WebSearch(), plugin_name="web")
kernel.add_plugin(Calculator(), plugin_name="math")

agent = ChatCompletionAgent(
    kernel=kernel,
    name="Assistant",
    instructions="Use tools to answer questions accurately.",
)

from semantic_kernel.agents import AgentGroupChat
from semantic_kernel.agents.strategies import (
    SequentialSelectionStrategy,
    KernelFunctionTerminationStrategy,
)

chat = AgentGroupChat(
    agents=[researcher, writer, editor],
    selection_strategy=SequentialSelectionStrategy(
        agents=[researcher, writer, editor]
    ),
    termination_strategy=KernelFunctionTerminationStrategy(
        agents=[editor],
        max_iterations=6,
    ),
)

await chat.add_chat_message(message="Write a blog post about WebAssembly in 2025")

async for response in chat.invoke():
    print(f"[{response.name}]: {response.content}")

from semantic_kernel.agents.strategies import SequentialSelectionStrategy

strategy = SequentialSelectionStrategy(
    agents=[researcher, writer, editor],
    initial_agent=researcher,
)

from semantic_kernel.agents.strategies import KernelFunctionSelectionStrategy

selector_fn = KernelFunction.from_prompt("""
Given the conversation, determine which agent should respond next.
Agents: Researcher, Writer, Editor.
If research is needed, pick Researcher.
If draft is ready for writing, pick Writer.
If draft needs review, pick Editor.
Respond with only the agent name.
""")

strategy = KernelFunctionSelectionStrategy(
    kernel=kernel,
    function=selector_fn,
    agents=[researcher, writer, editor],
)

import asyncio

async def parallel_analysis(topic: str):
    tasks = [
        researcher.invoke(f"Research {topic}"),
        competitor_analyst.invoke(f"Analyze competitors in {topic}"),
        trend_analyst.invoke(f"Identify trends in {topic}"),
    ]
    results = await asyncio.gather(*tasks)
    # Feed aggregated results to synthesizer agent
    summary = await synthesizer.invoke(
        f"Synthesize these analyses:\n" +
        "\n---\n".join(r.content for r in results)
    )
    return summary

using Microsoft.SemanticKernel;
using Microsoft.SemanticKernel.Agents;
using Microsoft.SemanticKernel.Agents.Chat;

var kernel = Kernel.CreateBuilder()
    .AddAzureOpenAIChatCompletion(deployment, endpoint, apiKey)
    .Build();

ChatCompletionAgent researcher = new()
{
    Kernel = kernel, Name = "Researcher",
    Instructions = "You are a research analyst. Find key facts and statistics."
};

ChatCompletionAgent writer = new()
{
    Kernel = kernel, Name = "Writer",
    Instructions = "You are a technical writer. Produce clear, engaging content."
};

AgentGroupChat chat = new(researcher, writer)
{
    ExecutionSettings = new()
    {
        SelectionStrategy = new SequentialSelectionStrategy { InitialAgent = researcher },
        TerminationStrategy = new MaxIterationTerminationStrategy(4)
    }
};

chat.AddChatMessage(new("Write about Kubernetes security best practices"));

await foreach (var msg in chat.InvokeAsync())
{
    Console.WriteLine($"[{msg.AuthorName}]: {msg.Content}");
}

from fastapi import FastAPI
from semantic_kernel.agents import ChatCompletionAgent

app = FastAPI()
# Initialize agents at startup

@app.post("/chat")
async def chat(message: str):
    response = await agent.invoke(message)
    return {"response": response.content}