Run scripted actions in a hosted browser session

Browserbase provides a hosted browser that can be controlled programmatically to load pages and carry out actions like clicking or filling in a form. A Beam agent starts a session when a trigger fires, such as a task that requires checking or updating something on a website without an API, runs the approved sequence of actions, and reads back the result. Any step that behaves unexpectedly, such as a page layout that doesn't match what the agent expects or a login that fails, is routed to a person to investigate before the task is retried.

Capture page content and screenshots on demand

A Beam agent can use a Browserbase session to capture the current state of a page, such as a screenshot or the visible text, as evidence that a task was completed or as input to a decision, for example confirming an order status shown only in a web portal. It reads this captured content and applies an approved rule, such as updating a record with the confirmed status. Where the captured content is unclear, contradicts existing data, or shows an error page, a person is asked to check the source directly.

Manage session state across multi-step tasks

Multi-step web tasks, such as navigating through several pages to reach a specific record, need the browser session to hold state, like being logged in, across each step. A Beam agent manages a Browserbase session through such a sequence, reading and acting on each page's content as it goes and applying the approved rule for what to do with the outcome, such as updating a record once the final page confirms it. If the session times out, hits an unexpected prompt, or requires a step outside what was approved, the agent stops and hands the task to a person.