Amazon Bedrock AgentCore Browser now supports OS Level Actions, extending AI agent automation beyond the browser's web layer (DOM/CDP) to the full OS desktop. This enables agents to interact with native dialogs, security prompts, certificate choosers, right-click context menus, and keyboard shortcuts that Playwright and CDP cannot reach. The feature works via the InvokeBrowser API using an action-screenshot-reaction loop: the agent dispatches an action (mouse click, drag, scroll, key press, shortcut), captures a full-desktop screenshot, sends it to a vision model, and decides the next step. Eight actions are supported across three categories: mouse control (mouseClick, mouseMove, mouseDrag, mouseScroll), keyboard input (keyType, keyPress, keyShortcut), and visual capture (screenshot). No additional setup is required for existing browser configurations.
