RIN runs open large language models directly on your phone. It works fully offline with no account, and nothing leaves your device unless you choose to connect a cloud model. Chats and memory are stored locally, never on a server.
- Chat offline with on-device models, or connect a cloud model (OpenAI, Anthropic, Gemini, OpenRouter, Ollama, or any OpenAI-compatible endpoint). API keys stay on the device.
- Send images to vision-capable models.
- Talk to it with on-device speech-to-text, and have it read replies aloud with local voices.
- Summon it from the power button as your device assistant, answering over any screen.
- Transcribe audio and translate text as one-shot tools from the drawer.
- Ground answers with web search (DuckDuckGo, a self-hosted SearXNG, or Ollama).
- Pull relevant pieces from your saved notes into context with on-device memory (RAG).
- Theme the whole UI live, with white, dark, or beige backgrounds, in 15 languages.
- Ghost mode pauses memory and stops saving chats for a private session.
Get the latest RIN.apk from the Releases page. Requires Android 12 (API 31) or newer. On first launch, open Settings, then Models, then Language models, and download a model to start.
cd Android/src
export JAVA_HOME=/path/to/jdk-21 # for example Android Studio's bundled JBR
export ANDROID_HOME=/path/to/android-sdk
./gradlew :app:assembleDebugRequires JDK 21 and the Android SDK with platform android-37.
LiteRT-LM and TensorFlow Lite for on-device inference, Gemma and other open models, sherpa-onnx with Piper voices for text-to-speech, Vosk for speech recognition, ONNX Runtime for the RAG embedder, and Jetpack Compose for the UI. Inspired by Google AI Edge Gallery.
MIT.