Handle asynchronous and long running agents with Amazon Bedrock AgentCore Runtime

Amazon Bedrock AgentCore Runtime can handle asynchronous processing and long running agents. Asynchronous tasks allow your agent to continue processing after responding to the client and handle long-running operations without blocking responses. With async processing, your agent can:

Key concepts

Asynchronous processing model

The Amazon Bedrock AgentCore SDK supports both synchronous and asynchronous processing through a unified API. This creates a flexible implementation pattern for both clients and agent developers. Agent clients can work with the same API without differentiating between synchronous and asynchronous on the client side. With the ability to invoke the same session across invocations, agent developers can reuse context and build upon this context incrementally without implementing complex task management logic.

Runtime session lifecycle management

Agent code communicates its processing status using the "/ping" endpoint health status. 200 HTTP Status response with payload {"status": "HealthyBusy"} indicates the agent is busy processing background tasks. {"status": "Healthy"} indicates it is idle (waiting for requests). A session in idle state for 15 minutes gets automatically terminated.

Implementing asynchronous tasks

To get started, install the bedrock-agentcore package:

pip install bedrock-agentcore

AgentCore SDK provides following options for integration asynchronous processing.

API based task management

To build interactive agents that perform asynchronous tasks, you need to call add_async_task when starting a task and complete_async_task when the task completes. The SDK handles task tracking and manages Ping status automatically.

# Start tracking a task manually
task_id = app.add_async_task("data_processing")

# Do work...

# Mark task as complete
app.complete_async_task(task_id)

Custom ping handler

You can implement your own custom ping handler to manage the Runtime Session's state. Your agent's health is reported through the /ping endpoint:

@app.ping
def custom_status():
    if system_busy():
        return PingStatus.HEALTHY_BUSY
    return PingStatus.HEALTHY

Status values:

• "Healthy": Ready for new work

• "HealthyBusy": Processing background task

Complete example

First, install the required package:

pip install strands-agents

Then, create a Python file with the following code:

import threading
import time
from strands import Agent, tool
from bedrock_agentcore.runtime import BedrockAgentCoreApp

# Initialize app with debug mode for task management
app = BedrockAgentCoreApp()

@tool
def start_background_task(duration: int = 5) -> str:
    """Start a simple background task that runs for specified duration."""
    # Start tracking the async task
    task_id = app.add_async_task("background_processing", {"duration": duration})
    
    # Run task in background thread
    def background_work():
        time.sleep(duration)  # Simulate work
        app.complete_async_task(task_id)  # Mark as complete
    
    threading.Thread(target=background_work, daemon=True).start()
    return f"Started background task (ID: {task_id}) for {duration} seconds. Agent status is now BUSY."

# Create agent with the tool
agent = Agent(tools=[start_background_task])

@app.entrypoint
def main(payload):
    """Main entrypoint - handles user messages."""
    user_message = payload.get("prompt", "Try: start_background_task(3)")
    return {"message": agent(user_message).message}

if __name__ == "__main__":
    print("🚀 Simple Async Strands Example")
    print("Test: curl -X POST http://localhost:8080/invocations -H 'Content-Type: application/json' -d '{\"prompt\": \"start a 3 second task\"}'")
    app.run()

This example demonstrates:

Common issues and solutions

Long-running agent gets terminated after 15 minutes

This can happen when the application is single threaded and the ping thread is blocked.