Users can now watch Perplexity AI control pages in Comet Browser

· 2 min read
Comet browser controls the web page
Comet browser controls the web page

Comet Browser, an experimental browser that has gained attention for its automation capabilities, has introduced a new feature allowing users to interact with a side chat bar to extract and direct Perplexity AI actions on the current web page. Previously, Comet operated mainly in headless mode, automating tabs and background tasks without providing users with real-time visual feedback. With this update, users can now watch as Perplexity controls and interacts with web pages, complete with visual indicators and animations that show exactly what actions are being taken, such as clicking or filling forms.

This functionality benefits anyone who relies on browser automation, especially power users automating repetitive workflows like data entry, publishing, or testing. By letting users observe the process in real time, the new mode addresses common concerns about trust and transparency in automation tools. It also gives users the ability to intervene if something goes wrong and to refine their automation scripts more efficiently.

The update unlocks a broader range of tasks, extending automation beyond background jobs to more interactive and visible browser sessions. Notably, Comet can now access and transfer context between multiple open tabs and directly control the foreground tab. This paves the way for advanced use cases that traditional MCPs or background automation cannot easily cover.

Comet remains in limited access, with no clear timeline for a public release. The feature, however, marks a step forward in browser automation and could accelerate adoption once available more widely. The company behind Comet has positioned itself as a pioneer in bringing interactive and transparent AI-driven automation to the browser, which fits their current strategy of targeting advanced users and developers looking to automate complex workflows that go beyond the reach of standard APIs or headless tools.