Introducing Amazon Bedrock AgentCore Browser Device

At AWS Summit New York Metropolis 2025, Amazon Internet Companies (AWS) introduced the preview of Amazon Bedrock AgentCore browser device, a totally managed, pre-built cloud-based browser. This device permits generative AI brokers to work together seamlessly with web sites. It addresses two elementary limitations: first, basis fashions (FMs) are skilled on massive however static datasets and wish dynamic entry to present info when API entry isn’t available; second, organizations face vital challenges when trying to scale internet automation with AI for enterprise use instances.

The event of agentic AI techniques is transferring towards purposes that may execute advanced, multistep duties. For these brokers to be efficient, they require entry to dynamic, real-time information, notably from web sites and internet purposes that don’t provide APIs or the place API integration can be advanced. Furthermore, as companies search to deploy AI-powered automation throughout their operations, they want options that may reliably scale with out the operational overhead of managing browser farms or fixing advanced concurrency points. The AgentCore Browser Device gives these capabilities, permitting brokers to carry out duties equivalent to automating analysis, streamlining operations, and interacting with web-based purposes—all with the scalability, reliability, and safety of the AWS Cloud infrastructure. By offering a totally managed cloud-based browser, AWS addresses the crucial want for enterprises to deploy AI automation at scale throughout hundreds of concurrent classes, supporting use instances from customer support automation to large-scale information assortment and evaluation, with out the standard complexity and useful resource constraints of self-managed browser automation frameworks.

On this publish, we introduce the newly introduced Amazon Bedrock AgentCore Browser Device. We discover why organizations want cloud-based browser automation and the restrictions it addresses for FMs that require real-time information entry. We discuss key use instances and the core capabilities of the AgentCore Browser Device. We stroll via how one can get began with the device.

Why do you want the cloud-based AgentCore Browser Device?

Conventional browser automation approaches have sometimes required vital infrastructure administration, safety issues, and improvement experience. The introduction of a totally managed, cloud-based browser automation answer addresses a number of crucial wants, together with simplified infrastructure administration, enterprise-grade safety, international availability and scaling, and price optimization. Organizations now not must provision, keep, and scale browser situations to assist their automation wants. AWS now handles the advanced infrastructure necessities, so builders can concentrate on constructing clever agent capabilities somewhat than managing browser farms. Cloud-based browser automation gives remoted execution environments with AWS safety controls, decreasing the chance of information exfiltration or unauthorized entry that may happen in much less managed environments. With a cloud-based browser, you’ll be able to instantaneously deploy browser situations throughout the worldwide infrastructure of AWS in order that browser automation can scale. By providing browser automation as a managed service, organizations can use a consumption-based pricing mannequin as a substitute of sustaining always-on infrastructure, which might considerably cut back prices for intermittent workloads.

Use instances for cloud-based browser automation

Dealing with repetitive internet duties: With the introduction of Amazon Bedrock AgentCore Browser Device, organizations can now implement subtle browser automation at scale. Cloud-based browser automation excels at minimizing guide execution of repetitive duties throughout internet interfaces. AI brokers can populate advanced internet types throughout a number of techniques, validate entries, and keep compliance with enterprise guidelines. Brokers can navigate to inner dashboards, extract crucial metrics, and compile studies with out human intervention. For organizations managing massive user-generated content material domains, brokers can help human moderators by prescreening content material throughout a number of internet interfaces.

AI powered analysis and intelligence gathering: With cloud-based browser automation, AI brokers turn into highly effective analysis assistants. They routinely observe associated web sites for pricing adjustments, new product launches, or content material updates with common monitoring. You need to use AI brokers to assemble and analyze client sentiment throughout varied internet boards, assessment websites, and social domains to tell product improvement. With the AgentCore Browser Device, you’ll be able to create automated techniques that commonly scan trusted info sources to maintain inner information bases present.

Complicated workflow automation throughout techniques: Many organizations function throughout quite a few internet purposes that lack built-in workflows. Use the AgentCore Browser Device to automate buyer setup throughout a number of software-as-a-service (SaaS) techniques when APIs are unavailable. This helps keep consistency and reduces error charges. You possibly can monitor provider portals, stock techniques, and logistics companies to keep up visibility throughout advanced provide chains. By automating account creation and permission settings throughout quite a few inner internet purposes, worker onboarding turns into streamlined.

Testing and high quality assurance: Cloud-based browser automation permits sturdy testing at scale. You need to use AgentCore Browser Device to validate person experiences and performance throughout completely different situations, gadgets, and browsers in parallel. Deploy brokers to constantly work together with crucial enterprise purposes and arrange alerts to your groups about efficiency points earlier than clients encounter them. With AgentCore Browser Device, you’ll be able to commonly take a look at internet purposes for accessibility compliance, safety vulnerabilities, or regulatory necessities.

Legacy system integration: Many organizations keep legacy techniques that lack trendy APIs. Allow trendy AI capabilities to work together with legacy internet purposes that might be pricey to interchange or modernize. Apply clever automation to techniques that had been by no means designed for programmatic entry. Because of this, you’ll be able to extract helpful organizational information trapped in older internet purposes via common, automated harvesting.

Core capabilities

The Amazon Bedrock AgentCore Browser Device empowers AI brokers to work together with internet content material the identical manner people do, via a totally managed distant browser infrastructure that minimizes conventional complexity whereas delivering enterprise-grade safety and scalability.

Internet interplay capabilities

• Full navigation management throughout web sites and multipage workflows
• Interplay with JavaScript-heavy purposes and dynamic content material
• Type manipulation, together with textual content fields, dropdown menus, and file uploads
• Humanlike interplay patterns equivalent to scrolling, hovering, and clicking

Serverless browser infrastructure

• Zero-management browser fleet with automated patching
• Seamless scaling from single session to hundreds primarily based on demand
• International deployment choices with usage-based pricing
• Optimized efficiency with out infrastructure overhead

Visible understanding

• Full-page screenshots enabling AI comprehension of structure and content material
• Visible factor identification by look and place
• Content material extraction from graphical components
• Decision and system emulation capabilities

Human-in-the-loop integration

• Actual-time interactive viewing and management for human operators
• Session recording for assessment, coaching, and compliance

Enterprise-grade safety

• Full session isolation for every browser occasion
• AWS Id and Entry Administration (IAM) controls for entry administration
• Ephemeral browser classes that reset after every use

Complicated internet software assist

• Full compatibility with trendy JavaScript frameworks
• Authentication dealing with and session persistence
• Processing of asynchronous content material and real-time updates
• Clever interplay with advanced UI patterns

Audit and compliance

• Detailed interplay logging and session recording
• Integration with AWS CloudTrail for complete monitoring

Observability

• Efficiency metrics on latency and useful resource utilization
• Integration with Amazon CloudWatch for unified monitoring
• Session file and replay for observability

This complete set of capabilities bridges the basic hole between AI brokers and the human internet, enabling organizations to construct clever brokers that may perceive and work together with content material designed for people somewhat than being restricted to API-based integrations.

How an AI agent can use AgentCore Browser Device

Amazon Bedrock AgentCore Browser runs in a safe, remoted containerized surroundings inside AgentCore, insulating internet exercise out of your native system. You possibly can work together with the AgentCore Browser Device utilizing browser actuation libraries, equivalent to Playwright, or use AI agentic frameworks specialised for browser automation, equivalent to Amazon Nova Act and Browser Use. It’s also possible to combine browser automation as a device in a multi-agentic workflow.

Amazon Nova Act or Browser Use works with the AgentCore Browser Device to take pure language directions from the person and convert them to actuations on the browser by following this workflow:

1. The person sends a question equivalent to “seek for sneakers on Amazon”
2. An agentic framework equivalent to Amazon Nova Act or Browser Use passes the question to the massive language mannequin (LLM)
3. The LLM causes and generates directions in a structured output format (for instance, JSON encoded)
4. The agentic framework maps these directions into browser actuation instructions (equivalent to Playwright, Puppeteer, or Selenium)
5. The browser actuation instructions are executed on the AgentCore Browser over a safe WebSocket connection
6. The response from the browser and a screenshot are despatched to the agent to cause additional

This course of repeats till the unique job is full. The circulation is illustrated within the following diagram.

Get began

The Amazon Bedrock AgentCore Browser Device is obtainable to be used at the moment. For a set of open supply examples, go to the amazon-bedrock-agentcore-samples repository on GitHub.

Conditions

To make use of the Amazon Bedrock AgentCore Brower Device, you could full the next conditions:

• Python 3.10+
• Confirm your IAM person or position has the permissions to make use of AgentCore Browser:

git clone https://github.com/awslabs/amazon-bedrock-agentcore-samples.git
pip set up bedrock-agentcore 

For browser visualization in your native machine, you want the BrowserViewerServer element within the repository you cloned at: 01-tutorials/05-AgentCore-tools/02-Agent-Core-browser-tool/interactive_tools

It’s also possible to visualize the browser reside on the Amazon Bedrock AgentCore console at https://us-east-1.console.aws.amazon.com/bedrock-agentcore/builtInTools

The next Python code demonstrates how one can use the AgentCore Browser Device instantly with the Playwright library and the Amazon Bedrock AgentCore SDK. This instance initiates a safe browser session, connects to it, and automates an easy workflow by which it navigates to https://www.amazon.com and searches for a product.

1. To get began with playwright:

cd 01-tutorials/05-AgentCore-tools/02-Agent-Core-browser-tool

2. Set up dependencies:

3. Writer your playwright-based script:

from playwright.sync_api import sync_playwright, Playwright, BrowserType
from bedrock_agentcore.instruments.browser_client import browser_session
from browser_viewer import BrowserViewerServer
import time
from wealthy.console import Console
console = Console()
def run(playwright: Playwright):
    # Create the browser session and maintain it alive
    with browser_session('us-west-2') as consumer:
        ws_url, headers = consumer.generate_ws_headers()
        # Begin viewer server
        viewer = BrowserViewerServer(consumer, port=8005)
        viewer_url = viewer.begin(open_browser=True)
        # Join utilizing headers
        chromium: BrowserType = playwright.chromium
        browser = chromium.connect_over_cdp(
            ws_url,
            headers=headers
        )
        context = browser.contexts[0]
        web page = context.pages[0]
        attempt:
            web page.goto("https://amazon.com/")
            console.print(web page.title())
            # Maintain working
            whereas True:
                time.sleep(120)
        besides KeyboardInterrupt:
            console.print("nn[yellow]Shutting down...[/yellow]")
            if 'consumer' in locals():
                consumer.cease()
                console.print("✅ Browser session terminated")
        besides Exception as e:
            console.print(f"n[red]Error: {e}[/red]")
            import traceback
            traceback.print_exc()
with sync_playwright() as playwright:
    run(playwright)

Alternatively, you’ll be able to construct a browser agent utilizing Amazon Nova Act to automate internet interactions:

4. Join Nova Act at https://nova.amazon.com/act and generate an API key.
5. In the identical Python digital surroundings:

pip set up nova-act

6. Writer your Nova Act primarily based script:

import time
from bedrock_agentcore.instruments.browser_client import browser_session
from nova_act import NovaAct
from wealthy.console import Console
from browser_viewer import BrowserViewerServer
 
NOVA_ACT_API_KEY = "YOUR_NOVA_ACT_API_KEY"
console = Console() 
 
def principal():
    attempt:
        # Step 1: Create browser session
        with browser_session('us-west-2') as consumer:
            print("r   ✅ Browser prepared!                    ")
            ws_url, headers = consumer.generate_ws_headers()
 
            # Step 2: Begin viewer server
            console.print("n[cyan]Step 3: Beginning viewer server...[/cyan]")
            viewer = BrowserViewerServer(consumer, port=8005)
            viewer_url = viewer.begin(open_browser=True)
 
            # Step 3: Use Nova Act to work together with the browser with NovaAct
            with NovaAct(
                    cdp_endpoint_url=ws_url,
                    cdp_headers=headers,
                    preview={"playwright_actuation": True},
                    nova_act_api_key=NOVA_ACT_API_KEY,
                    starting_page="https://www.amazon.com",
                ) as nova_act:
                    end result = nova_act.act("Seek for espresso maker and get the small print of the bottom priced one on the primary web page")
                    console.print(f"n[bold green]Nova Act Outcome:[/bold green] {end result}")
            
            # Maintain working
            whereas True:
                time.sleep(1)
             
    besides KeyboardInterrupt:
        console.print("nn[yellow]Shutting down...[/yellow]")
        if 'consumer' in locals():
            consumer.cease()
            print("✅ Browser session terminated")
    besides Exception as e:
        print(f"n[red]Error: {e}[/red]")
        import traceback
        traceback.print_exc()
 
if __name__ == "__main__":
    principal()

Alternatively, you’ll be able to run the tutorial notebooks within the Amazon Bedrock AgentCore GitHub repository.

Pricing and availability

Amazon Bedrock AgentCore presents versatile, consumption-based pricing with no upfront commitments or minimal charges. AgentCore Browser can be utilized independently of the opposite companies. You possibly can attempt AgentCore companies at no extra cost till September 16, 2025. After this date, AgentCore Browser Device shall be charged primarily based on consumption. Billing is calculated per second, utilizing the best watermark of CPU and reminiscence utilization for that second, with a 1-second minimal. 128 MB minimal reminiscence billing applies. Community information switch via buyer elastic community interfaces is billed at normal Amazon Elastic Compute Cloud (Amazon EC2) charges

For extra details about pricing, go to Amazon Bedrock AgentCore (Preview) Pricing.

Conclusion

Amazon Bedrock AgentCore Browser Device marks a transformative development in AI-powered internet automation, providing organizations a totally managed, cloud-based browser answer. AgentCore Browser Device addresses crucial limitations confronted by generative AI techniques requiring real-time information entry, enabling AI brokers to work together naturally with web sites via capabilities equivalent to full navigation management, visible understanding, and seamless integration with frameworks equivalent to Playwright and Amazon Nova Act. By utilizing this device, companies can now implement subtle automation at scale throughout varied use instances—from streamlining repetitive internet duties and conducting AI-enhanced analysis to automating advanced workflows and integrating with legacy techniques—all whereas benefiting from the dependable cloud infrastructure of AWS that adapts to organizational wants with out the operational overhead of managing browser farms.

Sources

To be taught extra and begin constructing, go to the next assets:

In regards to the authors

Veda Raman is a Senior Specialist Options Architect for generative AI and machine studying at AWS. Veda works with clients to assist them architect environment friendly, safe, and scalable machine studying purposes. Veda focuses on generative AI companies like Amazon Bedrock and Amazon SageMaker.

Rahul Sharma is a Senior Specialist Options Architect at AWS, serving to AWS clients construct and deploy, scalable Agentic AI options. Previous to becoming a member of AWS, Rahul spent greater than decade in technical consulting, engineering, and structure, serving to firms construct digital merchandise, powered by information and machine studying. In his free time, Rahul enjoys exploring cuisines, touring, studying books(biographies and humor) and binging on investigative documentaries, in no specific order.

Kishor Aher is a Principal Product Supervisor at AWS, main the Agentic AI crew accountable for creating first-party instruments equivalent to Browser Device, and Code Interpreter. As a founding member of Amazon Bedrock, he spearheaded the imaginative and prescient and profitable launch of the service, driving key options together with Converse API, Managed Mannequin Customization, and Mannequin Analysis capabilities. Kishor commonly shares his experience via talking engagements at AWS occasions, together with re:Invent and AWS Summits. Exterior of labor, he pursues his ardour for aviation as a common aviation pilot and enjoys enjoying volleyball.