Desert Ant Labs

Desert Ant Labs has developed a suite of specialized on-device micro-models, each designed to perform a single task: speech recognition (Voz), speech enhancement (Clear), sensitive information redaction (Redact), word-level timestamping (Align), filler word detection (Uhm), language identification (Tongue), title and description generation (Title), and topic tagging (Gist), among others. The models run directly on the user’s device, and developers can integrate them with just a few lines of code via the SDK—eliminating cloud-based costs charged per token.

The official example given is transcribing 10 minutes of audio in 2 seconds on an iPhone. It’s well-suited for app developers looking to add “small yet reliable” AI capabilities to their products, especially in scenarios where privacy is a priority—such as when developers don’t want to send user data to the cloud—or where high call volumes would make cloud API billing unaffordable. The core code is open-sourced on GitHub. It is not a general-purpose large model, and complex inference tasks are beyond its capabilities.