570775-ux / Benetto

What is Benetto?

Benetto is an Android voice notes app that transcribes speech entirely on your device—no cloud, no API calls, nothing leaves your phone. It uses Whisper.cpp for speech recognition and a Rust-based native layer for audio processing, wrapped in a clean Kotlin frontend with Jetpack Compose. The app records your voice, runs it through an on-device Whisper model, and stores the transcripts locally in a Room database. It also includes experimental LLM summarization using Qwen2.5 running via Candle.

Why is it gaining traction?

Privacy is the hook. Every other voice transcription tool sends your audio to servers—Benetto refuses to. The entire pipeline runs locally after the ~466MB Whisper Small model downloads once. For developers tired of API latency, costs, and privacy concerns, this is a refreshing architectural reference. The multilingual support (99+ languages) via Whisper adds broad appeal without any cloud dependency.

Who should use this?

Android developers building privacy-first features will find the JNI bridge and Rust integration patterns useful as a reference. Privacy-conscious power users who want voice notes without surveillance capitalism can use the APK directly. If you're evaluating on-device ML architectures for production apps, the Candle + Whisper combination here demonstrates a viable Kotlin/Rust stack.

Verdict

This shows promise but is early-stage—the Rust pipeline, audio recorder, and LLM engine are largely stubs with TODO comments. With a 0.8% credibility score and only 10 stars, treat this as a proof-of-concept reference rather than production-ready software. The architecture is sound; the implementation needs completion. Worth watching, not yet worth shipping.