AppRec lives in your Mac's menu bar. Pick one app, like Teams, Zoom or a browser tab, and it records that app's sound plus your microphone. When you stop, it writes a transcript and a summary, and names the file after what was said.
Everything runs on your Mac. No cloud, no account, no uploads. What AppRec connects to
Calls are private, so AppRec does all the work on your computer: recording, transcription and summaries. There's no cloud service and no account. This is every connection AppRec makes.
~/Music/Recordings and nowhere else.
localhost, which runs on your Mac, not on a server.
Stays on your MacUses the internet, never with your recordings
You do one thing: start and stop. Everything else happens on its own, in this order.
Only the app you picked is captured, so notification sounds and music from other apps stay out. Your microphone is mixed in.
A normal .m4a file lands in its own folder inside Recordings. Play it from the menu or open it anywhere.
Speech becomes text on your Mac. The language is detected for you, so English, Romanian and dozens more just work.
A short summary with key points and action items, and the file is renamed after the topic of the call.
One folder per recording, with three plain files inside. No account, no database, nothing locked inside the app.
Each summary has the same three parts: the gist, the key points, and who agreed to do what. It's written in the language people spoke in the call.
Login fix lands Wednesday, docs by Thursday, and dark mode moves to the next release.
Whisper large-v3 turbo runs on your Mac's Neural Engine and GPU. On an M3 Max, a 30 second clip takes about five seconds. It recognizes the spoken language by itself, so you never pick one.
roÎn Parlament, are la dispoziție 10 zile. În această perioadă, potrivit legii, poate să beneficieze de protecție SPP, dar și de un birou în care să lucreze.
enOkay, so for the release on Friday, Maria will finish the login fix by Wednesday. Tom, can you update the docs?
Instead of an app name and a timestamp, each recording's folder ends up with a short title from its summary. Your Recordings folder reads like a list of meetings.
Search your running apps by name and choose one. Calls, browser tabs, video players: if it makes sound, AppRec can record it. Turn the microphone off when you only want the app.
AppRec uses Whisper, OpenAI's open speech recognition model, in a version built for Apple chips. There's no server to set up: the model is a file on your Mac, and AppRec runs it there.
The first time a recording is transcribed, AppRec downloads Whisper large-v3 turbo (about 630 MB) from Hugging Face, where Argmax publishes it for Apple's Core ML. The menu shows the download progress. It's saved in ~/Library/Application Support/AppRec and never downloaded again.
macOS then optimizes the model for your chip's Neural Engine and GPU. This can take a few minutes, and the menu says so while it happens.
AppRec reads the recording, splits it at pauses, recognizes the spoken language and writes transcript.txt next to the audio. No internet connection is needed, and the audio never leaves your Mac.
One command with Homebrew, or build it from source with Xcode. AppRec isn't notarized by Apple yet, so Homebrew clears the macOS download warning for you.
The first time you record, macOS asks for permission to capture screen and system audio and to use the microphone. AppRec only captures sound.
brew install --cask hackmajoris/apps/apprecbrew upgrade --cask apprecNeeds Xcode 26 or later and takes a few minutes.
git clone https://github.com/hackmajoris/apprec
cd apprecbrew install xcodegenmake install