Amazon Bedrock AgentCore - Browser Tool | Amazon Web Services

Key Takeaways

Hello everyone. As part of the Amazon bedrock agent core video series, we are going to cover one of the two built-in tools in agent core as part of this video. So we have two built-in tools as part of the agent core. One the code interpreter, the other one is the browser tool. In this video we are going to cover the browser tool. Agent Core browser tool provides a comprehensive fully managed browser-based environment that enables agents to interact with web interfaces through natural language commands and automated actions. When you provide an agent with a complex query like buy shoes on Amazon, the system processes the request by first invoking the LLM to understand the intent. Then utilizing specialized browser automation tools that can translate highlevel instructions into specific browser commands and interactions. The platform provides a complete execution environment in the background including a headless browser instance and all the necessary infrastructure to support web automation tasks. This environment provides agent with the ability to navigate websites, fill forms, click buttons, and scroll pages, and perform other browser-based actions that would typically require human interaction. The system includes sophisticated computer use capabilities that allows agents to interact with web applications in a way that mimics human behavior while maintaining security and isolation. What makes this powerful is the visual feedback loop. After each action is executed in the browser environment, the system captures the screenshot of the current state and presents them back to the agent that allows the agents to observe the results of its actions, understand the current context of the web page and make the informed decision about the subsequent steps. This iterative process enables agent to handle complex multi-step web workflows. It's an ideal solution for automating task that requires sophisticated web interaction capabilities while providing a fully managed easy to implement experience for the developers. With this we'll go into the code walkthrough in the demo. This scripts creates an browser automation system by combining Amazon Bedrock agent course browser tool with Novaax natural language processing capabilities. It creates a browser session through agent core, connects Noahact to the browser via CDP, the Chrome dev tools protocol, and enable users to perform complex web tasks like searching and data extraction using simple natural language commands. You'll have to configure your no act key here. Then you execute the script by providing your natural language instructions, target website URL, and the Nova act API key as a command line arguments. The system will then display a detailed step-by-step execution log showing each browser action performed by the tool as it interprets and carries out your prompt. Next, we will see one more browser used to interact with agent core browser tool and view the browser live. This script creates a comprehensive browser automation system that combines agent core browser with a browser use library and provide live visual monitoring. It creates a browser session through agent core, starts a live viewer server on port 8000 for realtime browser viewing. creates a persistent browser session connected via CDP websocket and uses clot 3.5 sonnet to execute natural language browser task while allowing users to watch the automation happen in real time through a web interface with manual control capabilities.