Bubble

Modern apps are bloated with addictive features. Fed a never-ending stream of reels and shorts, today’s apps encourage teenagers to “doom-scroll,” resulting in a national uptick of 2 hours per day on the phone. And yet, at the core of these apps are still useful features, providing human connection and vital information.

So, what if there was a way to selectively block undesired content?

Bubble is the world’s first ML-powered packet-blocker for both web and iOS. Using a custom-trained machine learning model, Bubble drops unwanted packets of addictive content, leaving only what the user finds valuable.

Bubble achieves this by routing packets through a custom VPN proxy, where a custom-trained Random Forest model running on-device decides whether to discard or keep the contents of the packet based on its header, size, and content. Connections from Bubble to the public internet are encrypted using standard SSL. Bubble re-encrypts data sent to individual apps via its own SSL certificate.

There is no product like Bubble. By inspecting and re-encrypting packets, Bubble can analyze features like the path (e.g. instagram.com/reels) and size of a request, and extract labels from them. This packet-first approach makes Bubble generalizable to all apps, and robust against developer changes made at the API level. This packet-first approach also allows Bubble to target specific in-app functionality, an ability that modern app blockers are unable to do.

When we were building Bubble, the largest challenge was building an on-device VPN for iOS. There was no documentation on writing VPNs in Swift, and Arjun engineered the VPN from the protocol up.

We’re really excited to share Bubble, and we can’t wait for you to try it out.