Show HN: Offline AI assistant for Android (PDFs, Wikipedia, more) (github.com)

Eva, a new fully offline AI assistant for Android, has been launched, allowing users to access functionalities like document search, offline maps, music playback, and Wikipedia without any internet connection. This innovative tool integrates a language model, speech recognition, and text-to-speech capabilities directly on the device, ensuring that no data leaves the phone. Users can easily navigate through various features, including chatting, media playback, and document management, all organized through a user-friendly interface. The significance of Eva lies in its emphasis on data privacy and offline usability, making it a groundbreaking development for the AI/ML community. By leveraging advanced local processing, it enhances user experience and security, as sensitive information remains on the device. The application utilizes a combination of a sharded int8 vector index and full-text search for document retrieval, while cached map tiles and media are stored directly on the phone. Furthermore, it employs a flexible framework that allows users to select model types for different tasks—fostering adaptability across unique user needs—while maintaining streamlined storage management. This positions Eva as a prominent example of how AI can function effectively in resource-constrained environments.