European AI lab building small, opinionated on-device models.
Every model is free up to 100k monthly active devices per SDK. Unlimited inference per user.
Drop-in SDKs for iOS, Android, Web, and Node, with open weights on Hugging Face.
| Model | What it does | Platform | Docs |
|---|---|---|---|
| Align | Word-timestamp refinement for Apple's SpeechAnalyzer pipeline. | Apple | SDK Model |
| Clear | On-device speech enhancement: denoise, dereverb, and loudness-normalize. | Apple · Android · Linux · Windows · Web · Node | SDK Model |
| Clips | Short clips and highlights from talking video and audio: podcasts, interviews, meetings. On-device. | Apple · Linux · Windows | SDK Model |
| Ear | On-device spoken language identification across 99 languages. | Apple · Android · Linux · Windows · Web · Node | SDK Model |
| Emo | Multilingual on-device emoji suggestion. | Apple · Android · Linux · Windows · Web · Node | SDK Model |
| Gist | Multilingual on-device content topic tagging across a 36-topic taxonomy. | Apple · Android · Linux · Windows · Web · Node | SDK Model |
| Redact | Multilingual on-device PII detection and redaction. | Apple · Android · Linux · Windows · Web · Node | SDK Model |
| Shapes | On-device single-stroke shape recognition. | Apple · Android · Linux · Windows · Web · Node | SDK Model |
| Title | On-device titles and descriptions: a short factual title and a one- to two-sentence description for any passage of text. | Apple | SDK Model |
| Tongue | On-device language identification for short text across 84 languages. | Apple · Android · Linux · Windows · Web · Node | SDK Model |
| Uhm | On-device filler-word detection: frame-precise "uh"/"um"/"hmm" spans. | Apple | SDK Model |
| Voz | On-device speech recognition: transcripts with word-level timestamps, 25 languages. | Apple | SDK Model |
Weights exist and the models work, but no SDK ships them yet, so there is nothing to install today. Ask us if you want early access.
| Model | What it does | Docs |
|---|---|---|
| Eye | On-device frame scoring: which shot to keep from a burst or a clip. | Model |
| Face | On-device face matching across a photo library or through a video. | Model |
| Moderator | On-device NSFW image detection, trained only on licensed and synthetic data. | Model |
| Schemer | On-device structured extraction into a caller-supplied JSON schema. | Model |
| Toxic | On-device hate-speech triage for European languages. | Model |
| Who | On-device speaker labeling: per-person turns with timestamps. | Model |
- All SDKs on GitHub
- Models on Hugging Face
- Website
- License
- Commercial licensing: licensing@desertant.com
- Request early access: contact@desertant.com