Desert Ant Labs launches 18 specialised AI models for on-device use
The European lab says its audio, vision and text models run locally through Swift, Kotlin and JavaScript SDKs, with free use for up to 100,000 monthly active devices.
Desert Ant Labs has launched 18 specialised artificial-intelligence models designed to run directly on devices, targeting audio, vision and text tasks. The European lab said 12 models are stable and six are in beta.
The models are available through Swift, Kotlin and JavaScript software development kits, as well as the company’s website, GitHub and Hugging Face. Desert Ant Labs said use is free for up to 100,000 monthly active devices, although it did not specify the conditions that apply beyond that threshold.
The lab, created by the team behind the Detail video app, is positioning local inference as a way to reduce latency, operating costs and the amount of user data sent to external services. It said the models are small enough to run on some older phones and are intended to support frequent tasks such as transcription, audio enhancement and video processing.
Desert Ant Labs claims its local models can outperform cloud services and other models on speed, quality and cost. Those claims, along with statements about energy use, device compatibility and performance, have not been independently verified.
The company said Detail previously relied on cloud APIs for features including automatic editing and podcast audio enhancement. It plans for Detail 6 to replace those APIs with on-device models when the app launches with iOS 27, tying the timing and capabilities of that transition to Apple’s future software release.
The lab’s longer-term approach is to combine small local models with larger models or cloud services when a task requires them. It describes the local models as an always-on layer for routine work, with more capable systems used selectively.
