- 01
Install the build
Grab the latest Android package from the download page. This is internal testing — not on Play Store yet. Allow installs from trusted sources if Android prompts you.
- 02
Download a model
Open Models, pick a catalog entry or paste a compatible URL. LiteRT-LM uses .litertlm files (e.g. Gemma 4B); llama.rn loads .gguf via llama.cpp. Watch download progress, then load into memory.
- 03
Chat privately
The main Chat screen works like familiar assistants: streaming replies, stop/clear, a sidebar with saved threads, auto-titles, rename and delete. Text generation stays entirely on your device.
- 04
Build tools & actions
Tool Builder wizard creates reusable mini-tools backed by the loaded model. Mobile Actions translates plain language into structured JSON then Android intents — with confirmations before anything destructive runs.
- 05
Scan, speak, export
Document Scan uses ML Kit OCR on-device to feed scanned text into the model. Offline voice is powered by whisper.rn — choose model size in Settings. Export any assistant reply to a styled PDF.