desert-ant-labs/readable-runs
24 GB
None defined yet.
A European AI lab building opinionated on-device models and SDKs.
| Model | What it does | SDKs | Weights | Status |
|---|---|---|---|---|
| Align | Word-timestamp refinement for Apple's SpeechAnalyzer pipeline. | Align |
main | Stable |
| Clear | On-device speech enhancement: denoise, dereverb, and loudness-normalize. | Clear · ai.desertant:clear · @desert-ant-labs/clear |
v0.3.0 | Stable |
| Clips | On-device clip selection: a transcript's best non-overlapping moments, ranked. | Clips |
v0.1.0 | Stable |
| Emo | Multilingual on-device emoji suggestion. | Emo · ai.desertant:emo · @desert-ant-labs/emo |
v0.7.0 | Stable |
| Eye | On-device frame scoring: which shot to keep from a burst or a clip. | — | — | Closed beta |
| Face | On-device face matching across a photo library or through a video. | — | — | Closed beta |
| Gist | Multilingual on-device content topic tagging across a 36-topic taxonomy. | Gist · ai.desertant:gist · @desert-ant-labs/gist |
v2.2.0 | Stable |
| Moderator | On-device NSFW image detection, trained only on licensed and synthetic data. | — | main | Beta |
| Redact | Multilingual on-device PII detection and redaction. | Redact · ai.desertant:redact · @desert-ant-labs/redact |
v0.4.0 | Stable |
| Schemer | On-device structured extraction into a caller-supplied JSON schema. | — | main | Beta |
| Shapes | On-device single-stroke shape recognition. | Shapes · ai.desertant:shapes · @desert-ant-labs/shapes |
v0.3.0 | Stable |
| Title | On-device titles and descriptions: a short factual title and a one- to two-sentence description for any passage of text. | Title |
v0.1.0 | Stable |
| Tongue | On-device language identification for short text across 84 languages. | Tongue · ai.desertant:tongue · @desert-ant-labs/tongue |
Bundled · v1.0.0 | Stable |
| Toxic | On-device hate-speech triage for European languages. | — | main | Experimental |
| Uhm | On-device filler-word detection: frame-precise "uh"/"um"/"hmm" spans. | Uhm |
612592c | Stable |
| Who | On-device speaker labeling: per-person turns with timestamps. | — | — | Closed beta |
Every model is free up to 100k Monthly Active Devices. Unlimited inference per device.
Drop-in SDKs for iOS, Android, and Web ship with each available model, linked in the table above. More platforms and models are on the way.
Email contact@desertant.com for early access.
On-device multilingual PII redaction, in the browser.
Multilingual hate-speech triage scores, fully on device
Twelve real recordings, raw vs cleaned, on device
On-device content topic tagging across 101 languages
On-device language ID — a 2 MB model that runs in-browser