On-device AI Models for iOS and macOS
Drop-in Swift packages that run small, specialized models on device, on iOS and macOS, in a few lines of code on Core ML.
Every available Desert Ant Labs model ships as a Swift package built on Core ML, with async/await throughout and an API that reads like part of AVFoundation.
Add the package in Xcode and call the model. Everything runs on device, so audio, images, and text never leave the device.
Available models
Reversible PII redaction across 27 languages, from an 11.6MB Core ML model bundled in the app.
Emoji suggestions from text in under 2ms, small enough to run on every keystroke.
Content topics from any text across 101 languages, from a 74MB Core ML model, or a 15MB English-only build.
PencilKit snap-to-shape in one call, from a 0.2MB Core ML model, no download.
DeepFilterNet 3 speech enhancement on Core ML: 302x realtime on an iPhone 16 Pro, a 5-minute clip in 1s.
Frame-precise filler-word detection straight from the waveform, no transcript.
Language identification for short text across 84 languages, from a 2MB model bundled in the app.
On-device NSFW image moderation from a 9.7MB Core ML model, so the check runs before an upload leaves the phone.
Tighter word timings for any transcript, half the error, from a 0.7MB Core ML model. Apple Speech first.
Shorts and highlights from a transcript, ranked, a 25-minute video in 9s on Core ML.
Spoken language identification across 99 languages from 30 seconds of audio, on Core ML.
A title and a one or two sentence description for any passage, on Apple silicon through MLX.
Speech to text with word timestamps, 25 languages, entirely on the Neural Engine: 10 minutes of audio in 2s on an iPhone.
Available soon
Pricing
Each model is free up to 100k monthly active devices. Inference is unlimited.