Docs

Offline AI on your phone

How to install BOAR, ask it questions with no signal, add your own knowledge and models, and check how it performs on your phone.

BOAR is an open-source Android app that keeps a small AI model and a library of knowledge on your phone. After one download at setup, it answers with no signal at all: in flight, underground, on a mountain, at sea. Every answer shows the articles it came from, and every answer is measured.

These docs cover using it, adding to it, and checking it.

What BOAR is

  • An offline research companion. Ask it to explain, compare or work something out. It searches its library on the phone and a model on the phone writes the answer.
  • A library you can check. The built-in library has about 5,300 short Wikipedia-derived articles, and an optional pack adds the introductions of Wikipedia's ~50,000 Vital Articles. Answers list the articles they used.
  • Yours. MIT-licensed, no account, no API, no Google Play Services. Bring your own documents, your own model and your own packs.
  • Measured. Every answer records the model, the time to the first word, the speed and the memory used. The results are published, failures included.

What it isn't

  • Not a frontier model. The default model has 1.5 billion parameters. It's good at plain questions and slips on long syntheses; bigger models are slower. See Models.
  • Not a doctor or a rescue service. It can tell you what hypothermia looks like. Get real help as soon as you can reach it.
  • Not multilingual yet. The library and search are English. The app's menus are in English and Portuguese.

Where to go next

Edit this page on GitHub · Something wrong? Open an issue