[FEATURE] Add on-device macOS model support for AI Enhancement

[lnovelli]

Problem

I’d love it if you could add an option in the AI Enhancement section for using the on-device macOS LLM.

I’ve been testing Yap (https://github.com/FrigadeHQ/yap), which is very fast because it relies entirely on on-device models. However, it doesn’t offer other models such as Parakeet v3, which is significantly better than SpeechAnalyzer in my experience.

One particularly nice feature of Yap is transcript clean-up using the embedded macOS models. It requires no additional download, which is a big advantage compared with downloading (end executing) the 3.77 GB fluid-1 model, and it works quite well in the end.

Proposed solution

Could you please consider adding something similar? An option to use the built-in macOS LLM for transcript clean-up would provide a fast, privacy-friendly alternative without requiring users to download and run a large AI model.

In any case, BIG congratulations on developing this app! I really enjoy using it every day, Keep up the great work, you’re helping so many people type faster and better! :)

Thanks!!!

Alternatives considered

No response

[altic-dev]

Macos models was there earlier but it's pretty bad. So i had to remove it unfortunately.

[altic-dev]

Macos models was there earlier but it's pretty bad. So i had to remove it unfortunately.

[lnovelli]

Didn't know it was there before! In my experience with YAP, it's not performing so bad :) It correctly cleans up things like:

• spoken corrections: "let's meet at 4pm, ehm, no sorry at 5pm" --> "let's meet at 5pm"
• repetitions: "I, I don't like rain" --> "I don't like rain"
• wrong words corrections: "are their options to get there quicker" --> "Is there an option to get there quicker?"

I know there are a lot of other things/improvements under the radar, but if not too hard, it would be truly awesome to have this AI polishing option back ^^ do you think that is possible?

Big thanks