I built this because I do consulting work under NDA a lot of the times and maybe I'm too paranoid but I didn't feel comfortable using cloud transcription apps. Even the ones that promise not to store your data still process it on their servers and passed to openai or other. I wanted something where audio literally never leaves the device.
And yes there are some apps that do this already but what I wanted was to record audio locally, transcribes using whisper.rn (using react native) then I knew VERY little times I was going to read an hour long meeting what I mainly want is for the times my mind goes in the cloud and something being said I dont hear it or a date or something important or how a feature is suppose to work so chatting with a local on device llm in this case llama 3.2 3b or qwen3 4b for iphone 17s and ask simply about it and more in detail. Finally everything being stored encrypted on device storage.
Also thought about what happens if i decide to switch devices, so i built a simple export import system you put a password on the db and everything plus audios get ziped into a .viska file then you just put that on the new device and import it into the new phone where you have the app also.
*Stack:* React Native, Expo, whisper.rn, llama.rn
The performance is not bad at all specially for ios, tested on iphone 16 pro 2hr recordin in around 15 min. For android sucks but no gpu via llama.rn yet so it is slow but because android is more gracious with background processing you can just leave and come back later while the app is transcribing.
This is a One-time purchase ($9.99 iOS / $7.99 Android launch price). No subscription, no accounts, no telemetry.
I think this would be good for anyone who can't or won't use cloud transcription: lawyers, doctors, execs, journalists, or just privacy-conscious people.
Here is the links to the apps: iOS: https://apps.apple.com/us/app/viska-local-ai-meeting-notes/i... Android: https://play.google.com/store/apps/details?id=com.robkicksta...
Happy to answer questions about the on-device ML implementation or anything else.