Transcribed on the device
Speech becomes text on your iPhone or Mac, with speakers told apart. It works offline, and the audio never leaves the device.
MINUTES · LIVE · macOS · iOS
Every meeting, sealed into memory.
Record on your iPhone or Mac. Minutes transcribes on the device while you talk, then separates speakers after you stop — the audio never leaves it. Keep the meeting in your vault, use cloud AI for insights and summaries when you choose, and hand actions to AI.
HOW IT WORKS
The room stays private. The record endures.
Live text, then a finished record
Text appears while you talk. After recording, the archive brings together the transcript, insights, decisions and actions. With insights off, the summary waits until you request it.
Actions, handed to AI
Pass an action item to AI. Follow it in Flow, on the Lock Screen and in the Dynamic Island; the result lands in your Done by AI folder.
A record you own
Every meeting is a Markdown file in your vault — searchable, citable, and yours to take anywhere.
From the first word to the finished task.
Three moments from one meeting — recorded, sealed, and handed on. This illustrative meeting shows live transcription, the post-recording archive and an action handed to AI.
Sample meeting · illustrative content01 · IN THE ROOM
Text appears while you talk.
Speech becomes text on the device while you talk. Speakers are identified after recording stops. Turning insights off stops further live analysis; it does not recall text already sent.
Recording — the audio never leaves this device.
- Speaker · undecided
Let's walk through the launch timing again — last time we said early October.
- Speaker · undecided
Early October is tight. Tokyo would rather start with a small closed beta.
- Speaker · undecided
We'll put together the beta list and send it to you before Friday.
- Speaker · undecided
Should the beta ship with the Japanese interface?
- Speaker · undecided
Yes — Japanese is the precondition for this beta.
Tokyo starts with a small closed beta instead of a full launch in early October.
Send the beta list to the Tokyo team before Friday
Insights are off. The transcript keeps building on this device.
Insights
With insights on, transcript text goes to the cloud · the audio never leaves the device
02 · AFTER YOU STOP
Stop, and the record assembles itself.
After recording stops, speakers are separated on the device. Identify yourself or rename speakers in the archive. Five views preserve the meeting as a Markdown file in your vault.
Summary
- Timing: no full launch in early October — Tokyo starts with a small closed beta.
- Precondition: the beta ships with the Japanese interface.
- Owners: we prepare the beta list and send it before Friday.
Insights
- Tokyo's concern is localization, not features — the date moves only for the Japanese interface.
- The size of the beta is not settled yet. Confirm it at the next meeting.
Decisions
Tokyo starts with a small closed beta
Early October is too tight for a full launch; the beta validates the Japanese interface first.
Actions
Send the beta list to the Tokyo team
Speaker 1 · me · Friday
Draft a test checklist for the Japanese interface
Speaker 3
Transcript
- Speaker 1 · me
Let's walk through the launch timing again — last time we said early October.
- Tanaka
Early October is tight. Tokyo would rather start with a small closed beta.
- Speaker 1 · me
We'll put together the beta list and send it to you before Friday.
- Speaker 3
Should the beta ship with the Japanese interface?
- Tanaka
Yes — Japanese is the precondition for this beta.
Saved to your vault · Markdown
03 · AFTER THE MEETING
What was promised in the room, handed to AI.
Hand an action item to AI. It runs as a real conversation you can open and continue, shows its progress where you would look for it, and files the result next to the rest of your work.
Draft a test checklist for the Japanese interface
Speaker 3 · from “Tokyo beta timing”
- FlowHanded to AI
A Flow card appears the moment you tap it · Working · step 3/7
- Lock ScreenAI · Working
from “Tokyo beta timing” — on the Lock Screen and in the Dynamic Island
- NotificationAI finished one item
Draft a test checklist for the Japanese interface
- Done by AIJapanese interface beta checklist.md
Saved in Done by AI / 2026-09 · open it from the action item with “View result”
- Every screen shows Japanese copy, with no English fallback
- Dates, times and names read naturally in Japanese
- Testers can switch the language in Settings
On iPhone, progress shows on the Lock Screen and in the Dynamic Island, and a notification arrives when it is done. On Mac, follow it in Flow.
FAQ
QUESTIONS, ANSWERED
01Does the audio go to the cloud?
No. Transcription and speaker separation run on the device; audio is never uploaded. With insights on, transcript text is sent to cloud AI. Turning insights off stops further live analysis, but does not recall text already sent. Generating a summary sends the full transcript to cloud AI; with insights off, it waits until you request it manually.
02Does it work offline?
Transcription does. Insights and the summary use cloud AI, so they need a connection.
03What do I need before the first meeting?
An on-device model: about 680 MB on iPhone and up to 1.7 GB on Mac. You can start recording before it finishes downloading; the text fills in once it is ready.
04What if a call interrupts the recording?
Minutes picks the recording back up when the interruption ends. If iPhone will not let it restart in the background, you get one notification — tap it and recording continues.
05Which platforms does it run on?
Mac and iPhone today. Windows support is on the way.
06What languages are supported?
Transcription follows the on-device model; summaries and insights follow your selected output language.
A durable record, sealed into the same vault as the work it informs.
Keep the meeting. Not the surveillance.
