I built Orb because I wanted AI on my phone that still worked when the network disappeared.
Most AI apps are simple from the app side: send the prompt to a server, stream the response back, and keep the real complexity in the cloud. Orb is the opposite. It runs local models on Android after a one-time model download, so the app has to deal with the annoying mobile parts directly.
A few things stood out.
If your app needs a 1-3 GB model before it becomes useful, the download flow is not plumbing. It is onboarding. Users need to know:
This is also why I pulled part of the download work into a React Native native module. Large model files need progress, resume, and background-friendly behavior. A normal "just fetch it" flow is not enough.
A privacy app cannot just say "local-first" and hope people believe it.
The product has to make the boundary clear:
For Orb, the point is simple: once the model is downloaded, chat, image/document workflows, speech-to-text, text-to-speech, and local reasoning should not require an internet connection. Android hardware is all over the place.
Some devices can use GPU acceleration. Some Snapdragon devices may have experimental NPU/HTP paths. Many devices need CPU fallback. The same model can feel completely different across phones.
So the app cannot be marketed like every phone gets the same experience. The honest version is: use the best local path available, fall back safely, and keep model choices realistic.
The strongest use cases are not "replace ChatGPT." They are moments where sending data to a remote service feels wrong or impossible:
Local AI is weaker than cloud AI in many ways, but it wins when the private/offline boundary matters.
Orb is my attempt at making this feel normal on Android: local models on-device, image and document upload, speech-to-text, text-to-speech, no account, no subscription, no ads, and no cloud chat backend.
It is Google Play only for now:
https://play.google.com/store/apps/details?id=com.falaq.orb