Enhanced AI Agent Browsing with Bedrock AgentCore

Amazon Bedrock AgentCore Browser significantly upgrades its capabilities for AI agents interacting with the web, introducing three new features: proxy configuration, browser profiles, and browser extensions. These enhancements provide essential controls for agents operating in complex enterprise environments, moving beyond basic page navigation to support sophisticated, stateful, and secure web interactions.

The **proxy configuration** feature enables AI agents to route browser traffic through an organization’s own external proxy servers. This is critical for maintaining IP stability, which is vital for compliance and session continuity with sensitive applications like healthcare or financial portals that often validate sessions based on source IP addresses. It also allows agents to access internal web resources behind corporate firewalls and integrates seamlessly with existing network infrastructure. Technically, users specify `proxyConfiguration` when calling `StartBrowserSession`, optionally using AWS Secrets Manager for authentication credentials, and define `domainPatterns` or `bypass.domainPatterns` for granular traffic routing.

**Browser profiles** address the need for persistent session state. They allow agents to save and restore cookies and local storage across multiple browser sessions. This eliminates the repetitive and time-consuming process of re-authenticating or rebuilding context, drastically improving efficiency and reliability for authenticated workflows, e-commerce testing (e.g., maintaining shopping cart state), and multi-step user journeys. The profile lifecycle involves creating a profile with `create_browser_profile()`, saving session state with `save_browser_session_profile()`, and restoring it in new sessions via `profileConfiguration` in `start_browser_session()`.

Finally, **browser extensions** enable loading standard Chrome extensions into AgentCore Browser sessions. This provides immense flexibility for customizing browser behavior, from blocking ads and injecting authentication tokens for internal applications to modifying page content for optimized agent interaction. Extensions are packaged as ZIP files, uploaded to Amazon S3, and referenced by their S3 location during session startup. This extensibility, coupled with proxy logging, helps meet stringent compliance requirements like FedRAMP, HITRUST, and PCI by offering enhanced access control and audit capabilities. These features collectively empower developers to build more robust, secure, and adaptable AI agents for diverse web automation tasks.

(Source: https://aws.amazon.com/blogs/machine-learning/customize-ai-agent-browsing-with-proxies-profiles-and-extensions-in-amazon-bedrock-agentcore-browser/)