How to Intermediate · 4 min read

How to build multi-agent system with Semantic Kernel

Q: How to build multi-agent system with Semantic Kernel

Use semantic_kernel to create multiple AI agents by defining separate skills or services within a Kernel instance. Coordinate these agents by invoking their skills programmatically, enabling a multi-agent system that collaborates on complex workflows.

Quick answer

Use semantic_kernel to create multiple AI agents by defining separate skills or services within a Kernel instance. Coordinate these agents by invoking their skills programmatically, enabling a multi-agent system that collaborates on complex workflows.

PREREQUISITES

Python 3.8+
pip install semantic-kernel
OpenAI API key (free tier works)
pip install openai>=1.0

Setup

Install the semantic-kernel Python package and set your OpenAI API key as an environment variable.

Install Semantic Kernel: pip install semantic-kernel
Set environment variable: export OPENAI_API_KEY='your_api_key' (Linux/macOS) or setx OPENAI_API_KEY "your_api_key" (Windows)

bash

pip install semantic-kernel openai

Step by step

Create a Kernel instance, add multiple AI agents as services with distinct skills, and invoke them to collaborate on tasks.

python

import os
import semantic_kernel as sk
from semantic_kernel.connectors.ai.open_ai import OpenAIChatCompletion

# Initialize kernel
kernel = sk.Kernel()

# Add two AI agents with different skills
agent1 = OpenAIChatCompletion(
    service_id="agent1",
    api_key=os.environ["OPENAI_API_KEY"],
    ai_model_id="gpt-4o-mini"
)
agent2 = OpenAIChatCompletion(
    service_id="agent2",
    api_key=os.environ["OPENAI_API_KEY"],
    ai_model_id="gpt-4o-mini"
)

kernel.add_service(agent1)
kernel.add_service(agent2)

# Define simple skills for each agent
@kernel.skill("agent1", "greet")
def greet_agent1(input_text: str) -> str:
    return f"Agent1 says: Hello, {input_text}!"

@kernel.skill("agent2", "farewell")
def farewell_agent2(input_text: str) -> str:
    return f"Agent2 says: Goodbye, {input_text}!"

# Use the agents collaboratively
name = "Alice"
greeting = kernel.invoke("agent1.greet", name)
farewell = kernel.invoke("agent2.farewell", name)

print(greeting)  # Agent1 says: Hello, Alice!
print(farewell)  # Agent2 says: Goodbye, Alice!

output

Agent1 says: Hello, Alice!
Agent2 says: Goodbye, Alice!

Common variations

You can use asynchronous calls, stream responses, or switch to different AI models like gpt-4o or claude-3-5-sonnet-20241022. Also, agents can share context by passing outputs between them programmatically.

python

import asyncio

async def async_multi_agent():
    greeting = await kernel.invoke_async("agent1.greet", "Bob")
    farewell = await kernel.invoke_async("agent2.farewell", "Bob")
    print(greeting)
    print(farewell)

asyncio.run(async_multi_agent())

output

Agent1 says: Hello, Bob!
Agent2 says: Goodbye, Bob!

Troubleshooting

If you see authentication errors, verify your OPENAI_API_KEY environment variable is set correctly.
If skills are not found, ensure you registered them with the correct service_id and skill name.
For network issues, check your internet connection and API endpoint accessibility.

✅

Key Takeaways

Use Kernel to manage multiple AI agents as services with distinct skills.
Invoke agents programmatically to build collaborative multi-agent workflows.
Support async and streaming calls for efficient multi-agent interactions.

Verified 2026-04 · gpt-4o-mini, gpt-4o, claude-3-5-sonnet-20241022

Verify ↗