Desert Ant Labs has launched 18 opinionated on-device AI models for audio, vision, and text tasks. The European lab claims models run in milliseconds at no cost, fitting on five-year-old phones. Voz transcribes 10 minutes of audio in two seconds—4.7x faster than Whisper.

Clear enhances five-minute recordings in one second. Redact masks personal data in 27 languages without server upload. Tongue identifies 84 languages from three words using a 2MB model, scoring 0.933 accuracy.

The lab emphasizes data sovereignty, keeping intelligence on-device to avoid cloud dependencies and compelled disclosure. Built from five years developing the Detail video app, Desert Ant addresses gaps between research and implementation. Detail 6, launching with iOS 27, replaces all cloud APIs with on-device models.

Every model is free up to 100k monthly active devices with no tokens or logins.