Recommendation: Use the AgentCore CLI for new projects

The AgentCore CLI (@aws/agentcore-cli) is now the recommended way to create, develop, and deploy AI agents on Amazon Bedrock AgentCore. It offers broader framework support, local development with hot reload, built-in evaluations, gateway management, and more.

Get started: npm i @aws/agentcore-cli

See the Migration Guide for step-by-step instructions to migrate existing projects. The AgentCore CLI docs cover the full commands reference, supported frameworks, and configuration.

Session Management

Agent that maintains conversation state using session IDs.

Handler Code

# handler.py
from bedrock_agentcore.runtime import BedrockAgentCoreApp
from bedrock_agentcore.context import RequestContext

app = BedrockAgentCoreApp()

# Simple in-memory session storage (use database in production)
sessions = {}

@app.entrypoint
def chat_handler(payload, context: RequestContext):
    """Handle chat with session management"""
    session_id = context.session_id or "default"
    message = payload.get("message", "")

    # Initialize session if new
    if session_id not in sessions:
        sessions[session_id] = {
            "messages": [],
            "count": 0
        }

    # Add message to session
    sessions[session_id]["messages"].append(message)
    sessions[session_id]["count"] += 1

    # Generate response
    count = sessions[session_id]["count"]
    return {
        "response": f"Message {count}: You said '{message}'",
        "session_id": session_id,
        "message_count": count
    }

app.run()

Usage

CLI

agentcore configure --entrypoint handler.py
agentcore deploy

# Start conversation
agentcore invoke '{"message": "Hello"}' --session-id conv1

# Continue conversation
agentcore invoke '{"message": "How are you?"}' --session-id conv1

# Session id is automatically persisted and reused in .bedrock_agentcore.yaml
agentcore invoke '{"message": "Goodbye"}'

# Start a new conversation
agentcore invoke '{"message": "Hello"}' --session-id conv2