Bedrock AgentCore Starter Toolkit

Warnings

• gotcha For all new Bedrock AgentCore projects, AWS now officially recommends using the `@aws/agentcore-cli` (an npm-based CLI) instead of this Python toolkit. This Python starter toolkit is primarily intended for existing Python-based workflows. The `@aws/agentcore-cli` offers broader framework support and features for local development.
  Fix: For new projects, consider installing and using the `@aws/agentcore-cli` via `npm i @aws/agentcore-cli`. Refer to the official AgentCore CLI documentation for migration guides if you need to move an existing project.
• gotcha The Bedrock AgentCore Starter Toolkit is currently in 'public preview' status. This means that APIs and functionalities may change as the SDK is refined, potentially leading to breaking changes in future minor or patch releases.
  Fix: Review the `CHANGELOG.md` or release notes before updating to new versions, and thoroughly test your deployments after any updates.
• gotcha Successful deployment and invocation of agents require correctly configured AWS credentials and appropriate IAM permissions for creating roles, ECR repositories, and Bedrock AgentCore runtimes. Common issues arise from insufficient permissions or incorrectly set AWS regions.
  Fix: Ensure your AWS CLI is configured (`aws configure`) and the IAM role used for deployment has policies allowing `bedrock:*`, `iam:*`, `ecr:*`, and `lambda:*` actions, especially `bedrock-agentcore:*` for runtime operations. Specify the `--region` argument during `agentcore configure` if not using your default AWS region.

Install

• pip install bedrock-agentcore-starter-toolkit==0.3.5 Install latest version

Imports

• BedrockAgentCoreApp

  from bedrock_agentcore.runtime import BedrockAgentCoreApp

  This class is used to wrap your agent's Python logic, making it deployable to the Bedrock AgentCore Runtime. The starter toolkit CLI then interacts with this wrapped application.

Quickstart

This quickstart demonstrates how to define a simple AI agent using `bedrock_agentcore.runtime.BedrockAgentCoreApp` and then deploy it using the `agentcore` command-line interface provided by the `bedrock-agentcore-starter-toolkit`. The Python code defines the agent's logic, which the CLI toolkit then containerizes and deploys to the AWS Bedrock AgentCore Runtime. Ensure your AWS credentials are configured (e.g., using `aws configure`) and you have the necessary IAM permissions to create roles, Lambda functions, and use Bedrock AgentCore services.

import os
from bedrock_agentcore.runtime import BedrockAgentCoreApp

def my_agent_logic(request):
    """
    Processes an incoming request for the AI agent.
    """
    prompt = request.get("prompt", "Hello from AgentCore!")
    # Simulate agent processing or integrate with an LLM
    response_content = f"The agent received your prompt: '{prompt}'."
    return {"response": response_content}

app = BedrockAgentCoreApp()

@app.entrypoint
def production_agent(request):
    """
    The entrypoint function for the Bedrock AgentCore Runtime.
    This function wraps your core agent logic.
    """
    return my_agent_logic(request)

if __name__ == "__main__":
    # This block allows local testing of the agent logic.
    # For deployment, the `agentcore` CLI interacts with this file.
    print("Starting local AgentCore development server...")
    print("Access at http://localhost:8080/invocations")
    app.run()

# To deploy and invoke via the CLI, run these commands in your terminal:
# 1. Save the above Python code to a file, e.g., `my_agent.py`.
# 2. Configure your agent: `agentcore configure --entrypoint my_agent.py --name my-bedrock-agent --region us-east-1`
#    (Ensure AWS credentials are configured, e.g., via `aws configure`)
# 3. Launch the agent: `agentcore launch`
# 4. Invoke the deployed agent: `agentcore invoke '{"prompt": "Tell me a fun fact about Python."}'`