This week we fixed the worst part of Browserbase

Browserbase has significantly improved its session replay feature by transitioning from a model that captured DOM mutations to one that records actual screen content. Initially, the company used rrweb to replay user interactions by capturing DOM changes, which, although efficient on paper, proved unreliable due to web complexity and inconsistencies in HTML and CSS representation. The team realized that recording the screen directly was more reliable, leading to a new system that captures timestamped PNG screenshots using the Chrome DevTools Protocol. These are then converted into video segments asynchronously, ensuring efficient resource usage and maintaining high fidelity without taxing the browser's CPU. This method allows for parallel encoding, making even lengthy sessions quickly accessible, with just-in-time encoding for unwatched segments to optimize costs. By adopting this approach, Browserbase has resolved previous issues, such as incorrect replays and complex multi-tab behavior, thus enhancing reliability and user trust in their session recordings.