Overview & result
Desert Ant Labs demonstrated an audio-file workflow: Ear detects the language, Voz transcribes, Redact removes personal information and Jev makes about 20 decisions in one call to choose what to do next. The post describes local audio models but does not say Jev inference itself runs on-device.
From the original post
Jev + on-device models = results in seconds with no LLM in the loop.
Quick demo app to show the possibilities. Drop in an audio file: Ear detects the language, Voz transcribes it and Redact removes PII.
Then @typesafeai's Jev makes about 20 decisions in one call in milliseconds, and picks which of our on-device models to run. Voice memo to to-do list. Meeting to redacted transcript. Podcast to clips.
How Jev fits in the loop
- Drop an audio file into the demo.
- Ear detects language; Voz transcribes; Redact removes personal information.
- Jev evaluates about 20 decisions in one call and selects the next task or model.
How to reproduce
- Watch the linked demo and inspect any implementation details published by Desert Ant Labs.
- No public repository or benchmark procedure is linked from this case.
Why this build matters
It demonstrates how local audio processing can feed a compact decision step for different downstream tasks.
Limitations
- Local processing of the audio models does not imply local Jev inference.