amazon-bedrock-agentcore-samples/01-tutorials/01-AgentCore-runtime/01-hosting-agent/05-semantic-kernel-with-bedrock-model

Hosting Semantic Kernel Agents with Amazon Bedrock models in Amazon Bedrock AgentCore Runtime

This tutorial demonstrates how to deploy a Microsoft Semantic Kernel agent using Amazon Bedrock models to Amazon Bedrock AgentCore Runtime.

Overview

In this tutorial, you will learn how to:

• Create a Semantic Kernel agent with custom plugins
• Use function calling with Semantic Kernel and Amazon Bedrock
• Deploy your agent to AgentCore Runtime
• Invoke your deployed agent using both the toolkit and boto3

Prerequisites

Before starting this tutorial, ensure you have:

• Python 3.10+
• AWS credentials configured
• Docker or Finch running
• Access to Amazon Bedrock models (Claude Sonnet)

Quick Start

1. Install dependencies:

uv add -r requirements.txt --active

2. Run the agent locally:

python semantic_kernel_bedrock.py '{"prompt": "What is the weather now?"}'

3. Deploy to AgentCore Runtime: Follow the steps in the Jupyter notebook runtime_with_semantic_kernel_and_bedrock_models.ipynb

Agent Features

The example agent includes three tools:

• Calculator: Performs mathematical calculations using safe eval
• Weather: Returns weather information (dummy implementation)
• Get Time: Returns the current timestamp

Architecture

The agent uses:

• Framework: Microsoft Semantic Kernel
• Model: Amazon Bedrock (Claude Sonnet 3)
• Runtime: Amazon Bedrock AgentCore Runtime
• Tools: Custom plugins with function calling

Files

• semantic_kernel_bedrock.py: The main agent implementation
• runtime_with_semantic_kernel_and_bedrock_models.ipynb: Step-by-step tutorial notebook
• requirements.txt: Python dependencies
• images/: Architecture diagrams

Key Concepts

Semantic Kernel Plugins

Plugins in Semantic Kernel are classes that contain kernel functions. Each function is decorated with @kernel_function and can be automatically discovered and used by the AI model through function calling.

Function Calling

The agent uses Semantic Kernel's built-in function calling capabilities with Amazon Bedrock. When configured with FunctionChoiceBehavior.Auto(), the model can automatically decide when to call functions based on the user's input.

AgentCore Runtime Integration

The agent is deployed as a containerized service on AgentCore Runtime, providing:

• Serverless execution
• Automatic scaling
• Built-in monitoring
• Secure execution environment

Next Steps

After completing this tutorial, you can:

• Add more sophisticated plugins (e.g., database access, API calls)
• Experiment with different Amazon Bedrock models
• Integrate with other AgentCore capabilities (Memory, Gateway, Tools)
• Build multi-agent systems using Semantic Kernel

Troubleshooting

If you encounter issues:

1. Ensure Docker/Finch is running
2. Verify AWS credentials are configured
3. Check that you have access to the specified Bedrock model
4. Review the agent logs in CloudWatch

Related Tutorials

• Strands Agents with Amazon Bedrock
• LangGraph with Amazon Bedrock
• CrewAI with Amazon Bedrock