Blog · 28 July 2026
The most reliable way to capture what just happened in a meeting is to talk about it for ninety seconds while walking to the next one. The unreliable part is what happens to that recording afterwards.
Voice memos pile up. They are quick to make and slow to use, because listening back takes as long as recording and you cannot skim audio. So they sit there, and the actions inside them quietly expire.
The useful version of this is not transcription. It is transcription plus extraction: turn the audio into text, then pull out the things that are actually tasks and leave the rest behind.
You say something like:
"Okay so that was the Kaya call. They want the revised quote by Thursday, and they need it to include the installation cost separately this time. Ayşe is going to send over the floor plan, I should chase her Wednesday if it hasn't arrived. Oh and they asked about the warranty, I said I'd check and come back."
And you get back:
The rambling is gone, the ordering is preserved, and the deadlines came through. That is a rough note turned into something you can act on without listening to yourself talk for ninety seconds.
Meeting notes are one of the most sensitive things most people produce. They contain client names, prices, complaints about colleagues, deal terms and unformed opinions. They are also exactly the kind of content people upload to transcription services without a second thought.
When the transcription and the extraction both happen on the device, the recording never leaves the phone. Two practical consequences:
The quality of the list depends far more on how you talk than on the settings:
Being straight about the limits:
Once the list exists as text, it is just text. It can be dropped into Reminders, sent as a message, or pushed into whatever you actually use — and that step can be automated so the whole thing is one tap. Private AI inside Apple Shortcuts covers how that fits together.
An iPhone on iOS 17 or later and a one-off setup on Wi-Fi. After that it works with no connection. See will it run on my iPhone if you are on an older device.
Yes. If both the transcription and the extraction run on the device, a recording can be turned into a list of tasks with no connection at all. Setup needs Wi-Fi once; after that it works offline.
With an on-device assistant, no. The audio is transcribed on the phone and the tasks are extracted on the phone, which is why it works with no signal.
Good for clear speech in normal background noise, and it reliably picks up deadlines and names you say out loud. Unusual names are often spelled phonetically and need a quick correction. It will not invent tasks you did not mention.
Under a couple of minutes works best. Two short notes give better results than one long, rambling one.
OnDevice LLM is a private AI assistant that runs entirely on your iPhone — no account, no cloud, and nothing you type leaves the device. Free in full for three days, then a one-time purchase.
Last updated: 28 July 2026