appless

AppLess is a mobile app experiment built around one idea: a phone with no installed apps, where every screen is generated on the spot by an AI model when you ask for it. Instead of opening a weather app or a booking app, someone opens AppLess, types or speaks what they want, and a real native looking screen for it appears immediately, styled like iOS on iPhone and like Material Design on Android. Tapping something on that screen generates the next screen in the same way.

This is a demo of what the developers call a generative UI operating system, built on top of their own project called OpenUI. Rather than the AI returning a block of JSON or a fixed template, it writes a compact custom language that a renderer turns into live, interactive native components as the response streams in. The README is upfront that this is an experiment, not a working replacement for real apps. Screen content can be grounded in real information if optional web search and image tools are turned on, but actions like placing an order, booking a flight, or sending a payment are always simulated and do not do anything real.

Under the hood, one file defines every UI component the model is allowed to use, its properties, and a description of what it does. That file automatically generates the instructions given to the AI model, so the model always knows exactly what building blocks it can use to draw a screen. Two separate design systems then take that same generated screen and render it differently depending on whether the app is running on iOS or Android.

To run it, someone clones the repository, installs dependencies with npm, and starts it for iOS, Android, or Expo Go. On first launch it asks for a free API key from Cerebras, the company whose AI model actually generates the screens, and stores that key on the device. There is no backend server to set up. The project is licensed under MIT and openly invites people to fork it and wire in real services like food delivery or calendar APIs.