Minute records and summarizes meetings on your Mac without sending a bot to the call
The open-source app keeps notes as local files and separates meeting detection from actually opening the microphone.

The meeting recorder doesn’t need a seat in the meeting. Minute records on the Mac, runs its speech and language models locally and stores the resulting notes as files. No extra attendee has to arrive with a name that makes everyone wonder who invited the minutes.
The project README says “Minute is free and open source” and describes operation without an account or analytics. Its key architectural claim is local processing: recording, transcription, summaries and questions about a meeting run on the user’s machine. This account comes from the project’s documentation, rather than an independent network audit.
Meeting detection is a separate feature and is off by default. When enabled, the app checks which programs are running and using the microphone, then can offer to record. The documentation says detection itself doesn’t open the microphone or listen. Each supported app can be set to ask, record automatically or do nothing.
That separation gives the user a useful choice. A notification that a call began can be helpful without silently making every call into an audio archive. Minute also documents automatic stopping behavior around silence, closing phrases and an application quitting, with a way to keep recording when the meeting isn’t actually over.
Recording the other participants requires system-audio capture. The README specifies macOS 13 or later and Screen Recording permission for that feature; without it, the app records only the microphone. A transcript missing half the conversation may therefore be a permission problem rather than a remarkably quiet team.
After recording, the app produces a summary, decisions and action items above the transcript. It can answer questions using that meeting’s transcript and link an answer back to a time in the audio. That playback link is particularly useful when a polished summary sounds a little more decisive than the people in the call actually were (a meeting tradition).
Long recordings have a stated limit: if the meeting exceeds the model’s capacity, the summary says it covers only part of it. A partial summary needs to stay partial in the reader’s mind. The action list may look complete even when the source material being summarized wasn’t.
Minute also supports speaker labels and optional voice profiles, according to the README. Those features can make a note easier to navigate, but names and decisions still deserve a comparison with the recording before becoming somebody’s assigned work. Local processing changes where information goes; it doesn’t make generated text infallible.
For teams wanting fewer third-party services inside a call, this is an inspectable alternative with a concrete storage model. Its usefulness on a particular Mac depends on capture permissions, model choice and whether the machine keeps up with the actual meeting. The README offers the pieces; the user’s next recording supplies the test.
The bot can miss this meeting.