AWS Builder Center

Builder Center

Automating Web Interactions with Amazon Bedrock AgentCore Browser and Nova Act

Automating Web Interactions with Amazon Bedrock AgentCore Browser and Nova Act

Learn how Amazon Bedrock AgentCore Browser and Nova Act enable secure, AI-powered web automation through natural language prompts. This guide walks through a practical example of automated product search, explaining how AI agents can navigate websites, extract information, and perform complex tasks in isolated cloud environments without infrastructure overhead.

Makendran

Introduction

The Amazon Bedrock AgentCore Browser represents a significant advancement in cloud-based web automation, offering developers a secure, containerized environment for web interactions. Combined with Amazon Nova Act, it enables AI agents to navigate websites, extract information, and perform complex tasks with minimal infrastructure overhead.

What is Amazon Bedrock AgentCore Browser?

The AgentCore Browser is a secure, cloud-based browser environment that runs within Amazon Bedrock's AgentCore service. It provides isolation from local systems while enabling sophisticated web interactions. Key capabilities include:
  • Web Navigation: Agents can navigate websites, fill forms, click buttons, and parse dynamic content
  • Serverless Infrastructure: Automatically scales without requiring infrastructure management
  • Visual Understanding: Takes screenshots that allow AI agents to interpret websites as humans do
  • Session Isolation: Each browser session runs in a separate containerized environment
  • Observability: Real-time visibility through live viewing and session replay capabilities
  • Security: Built-in CloudTrail logging and audit capabilities

Introducing Amazon Nova Act

Amazon Nova Act is an AI-powered browser automation framework that works seamlessly with AgentCore Browser. Currently available in US regions only, Nova Act enables natural language-driven web interactions, allowing developers to automate complex web tasks with simple prompts.

Practical Example: Product Search Automation

Let's walk through a real-world example of using Nova Act with AgentCore Browser to search for products on Amazon and extract product details.

Setup

First, obtain your Nova Act API key from the Amazon Nova Act Homepage (US users only). Once you have your key, export it to your environment:
Bash
export NOVA_ACT_API_KEY="your-api-key-here"


Running the Browser Automation

The following script demonstrates searching for MacBooks and extracting details of the first result:
Bash
cd ~/workshop/agentcore/browser/
uv run nova_act_browser.py \
    --prompt "Search for macbooks and extract the details of the first one" \
    --starting-page "https://www.amazon.com/"


What Happens Behind the Scenes?

The automation follows a sophisticated multi-step process:
  1. Session Initialization: A browser session is created through Amazon Bedrock AgentCore's browser tool
  1. Natural Language Processing: Nova Act interprets the natural language prompt
  1. Action Planning: The AI agent analyzes the webpage and plans necessary actions
  1. Execution Steps:
  • Opens Amazon homepage
  • Locates and types "macbooks" into the search field
  • Clicks the search button
  • Identifies the first product result
  • Clicks on the first result
  • Extracts detailed product information

Step-by-Step Execution

Here's what Nova Act does at each step:
Step 1: The agent analyzes the homepage and identifies an empty search field
Plain text
think("I am on the amazon homepage. My task is to search for macbooks...")
agentType("macbooks", "<box>11,403,49,990</box>")

Step 2: After entering the search term, it locates and clicks the search button
Plain text
think("The search field now contains 'macbooks'... I need to click the search button...")
agentClick("<box>9,990,49,1034</box>")

Step 3: On the results page, it identifies and clicks the first product
Plain text
think("I am now on the search results page... I need to click on the first result...")
agentClick("<box>582,589,632,1409</box>")

Step 4: Finally, it extracts the complete product details
Plain text
return "Apple 2025 MacBook Air 13-inch Laptop with M4 chip: Built for Apple Intelligence, 13.6-inch Liquid Retina Display, 16GB Unified Memory, 256GB SSD Storage, 12MP Center Stage Camera, Touch ID, Midnight"


Viewing Results

Nova Act generates an HTML file containing a complete visual record of the automation run. This file shows screenshots of each step, making it easy to debug and verify the agent's actions. The output includes detailed metadata such as session ID, execution time, and number of steps executed.

Key Benefits

Security and Isolation: Every web interaction occurs in a separate containerized environment, protecting your local system from potential threats.
Scalability: The serverless infrastructure automatically handles scaling without requiring manual intervention or infrastructure management.
Transparency: Real-time visibility and session replay capabilities provide complete observability into agent actions.
Natural Language Interface: Developers can describe desired outcomes in plain English rather than writing complex automation scripts.
Complex Navigation: Handles sophisticated web applications that require full browser capabilities, including JavaScript-heavy single-page applications.

Use Cases

The AgentCore Browser with Nova Act is ideal for:
  • Automated testing of web applications
  • Data extraction from dynamic websites
  • Form filling and submission workflows
  • Competitive intelligence gathering
  • Price monitoring and comparison
  • Content verification and compliance checking

Conclusion

The Amazon Bedrock AgentCore Browser, combined with Amazon Nova Act, provides a powerful platform for AI-driven web automation. By combining natural language understanding with secure, scalable browser infrastructure, it enables developers to build sophisticated automation workflows without wrestling with infrastructure complexity or security concerns. As web interactions become increasingly central to AI agent capabilities, tools like AgentCore Browser are positioning themselves as essential components of the modern AI development toolkit.
# amazon-bedrock-agentcore
Any opinions in this article are those of the individual author and may not reflect the opinions of AWS.
Enjoyed reading this content? Let the author know!

Your likes, comments, shares, and saves help creators reach more builders.

Loading recommendations

Loading article