Prototype - Transcribe and process text entirely locally on your computer

I’ve for a long time been building my own tools that make everyday life easier. Before AI, it was often automations that could handle parts of my work that weren’t so… let’s say, super fun. Since Claude Code came into my life about 9 months ago, this has escalated.

The custom-built tool I by far use the most is “Transkribera”. It gets a lot of use a couple of times per day on average.
It’s an app for MacOS that transcribes meetings and thoughts entirely locally and securely on my computer. It also uses local language models to summarize the transcriptions.

Also… given the geopolitical situation we’re in today, it feels good to be able to handle both data and AI in a secure way that doesn’t leave my computer unless I want it to.

There’s also the option to add an API key in the app for Berget AI, a Swedish company that provides open language models. Then you can use their AI models. It won’t be local on your computer anymore, but no data leaves Sweden and you can also take advantage of their more powerful models.

For those of you who’ve followed me for a while, this will surely sound familiar. This is something I’ve explored, among other things with my colleagues Carl Heath and Björn Flintberg in various forms, in the prototypes Diane from early 2023 and Ragnarli a little later.
What has happened in recent years is that we’ve got really sharp transcription models in the form of KB Whisper (from Kungliga Biblioteket) as well as smaller language models that can run on ordinary computers.

I’ve also converted the KB Whisper models to CoreML so they can take even better advantage of the computer’s resources.
You can find those models at https://huggingface.co/mickekringai/kb-whisper-coreml

In the video I go through how the app works, if anyone else is keen to try it. You can transcribe by loading an audio/video file or by recording directly in the app.

Test the app?

The app isn’t released on the App Store because I don’t have time to support a free app, but if you want, you can either download the entire codebase from Github and build it yourself with Xcode. Or fork it.

The app is released as open source under the MIT license. https://github.com/mickekring/Transcribe-MacOS-App

There is also a pre-packaged version of the app at https://github.com/mickekring/Transcribe-MacOS-App/blob/main/Transcribe_v2.0(1).zip

That file can be downloaded, unpacked and dragged to “Applications”. Then it’s just to run.

NOTE! As usual when downloading things from the internet, make sure to scan for viruses and the like.