Two Things That Should Have Been Automatic

The best features are the ones you never notice. They just make the annoying thing stop being annoying. The latest s1ft release has two of those, and they share a theme: work that used to be manual now happens on its own, without any of your data leaving your Mac.

Your meetings name themselves now

Here’s a small friction I lived with for months. A call ends, s1ft has a clean recording and a full transcript — filed under a name like “zoom.us” or “Google Chrome.” Useful later only if I stopped and renamed it. So most of the time I didn’t, and my recordings folder turned into a wall of platform names I couldn’t tell apart.

The fix was sitting right there the whole time: my calendar already knew what every one of those meetings was called. “Weekly Standup.” “Acme — Contract Review.” “1:1 with Sarah.” The title I needed was in the event I’d already accepted.

So now s1ft reads it. Connect the calendar accounts you already have on your Mac — Google, Outlook, iCloud, whatever you’ve added in System Settings → Internet Accounts — and after each recording finishes transcribing, s1ft matches it to the calendar event that was happening at that time and renames the meeting to the event’s title. Back-to-back calls, overlapping invites, an all-day event sitting on top of a real meeting — when more than one event could match, it reads the transcript and picks the title that actually fits what was said, right there on your machine.

A few things I was careful about:

For naming, it uses event titles, times, and your own response. s1ft looks at each event’s title and start and end times, and at whether you’ve declined it, so it can skip meetings you’re not going to. To recognize a meeting in the first place, it also reads a little of the guest list: how many people were invited, the work email domains they belong to (free ones like Gmail are skipped), and whether you’re the organizer. It doesn’t use guest names or addresses for naming. It reads all of this on your Mac, through the macOS Calendar. Your calendar data stays on your Mac. There is no server to trust. s1ft never stores a calendar password. macOS owns the connection; s1ft just borrows the event. One more thing worth knowing: once a recording is named, the name is part of the meeting details s1ft includes when you ask for a summary. If you’ve set up a cloud provider such as Anthropic or OpenAI with your own key, that name goes to them, along with the transcript, under their terms. With a local engine like Ollama or LM Studio on your Mac, it stays on your machine. If you don’t use summaries, it goes nowhere.
You’re never locked out of your own names. If you’ve already renamed a meeting yourself, that name wins — always. Auto-naming only ever fills in the blanks it left behind.
It can’t break a recording. Naming runs after transcription, as a separate step. If calendar access is off, or nothing matches, the recording keeps its default name and everything else works exactly as before.

You turn it on per account in Settings → Calendar accounts, and you can flip any account off whenever you want. That’s the whole setup.

Your transcripts know who said what

The other change shows up inside the transcript itself.

Until now, everyone on the far end of a call landed in the transcript as one undifferentiated voice. Fine for a 1:1. Useless for a four-person design review where the whole point is who pushed back on what.

Now s1ft separates the voices. Your own audio comes in on your microphone and is labeled Me. When there are two or more distinct people on the other end, they’re split into Participant 1, Participant 2, and so on — so the transcript reads like a real conversation, and your summaries can point to who raised which concern.

I want to be precise about what this is, because the honest version is better than the hype:

It labels speakers. It doesn’t name them. s1ft can tell that three different people spoke; it can’t tell you that Participant 2 is Sarah. There’s no address book, no “who’s in the room” guesswork — just clean, numbered separation of distinct voices.
It forgets on purpose. The voice patterns it uses to tell speakers apart are computed for that one meeting and thrown away the moment it’s done. Nothing is stored, nothing is enrolled, and Participant 1 in today’s call has no connection to Participant 1 in yesterday’s. There is no voiceprint of you or anyone else sitting on a disk somewhere.
It runs entirely on your Mac. The separation happens on-device using Apple’s Neural Engine. No audio and no derived voice data ever leaves your machine — same promise as transcription itself.

There’s nothing to switch on. It just happens while your meeting transcribes, and if a call only has one other voice, you won’t see numbered participants at all — it quietly stays out of the way.

Why both of these look the same from the outside

Neither of these is a headline feature. You won’t open s1ft and see a new button. You’ll just notice, a week in, that your recordings folder finally makes sense at a glance and your transcripts read like conversations instead of monologues — and that you never had to send a calendar invite or a voice sample to anyone’s cloud to get there.

That’s the bar I’m trying to hold s1ft to. The useful thing should happen on its own, and it should happen on your machine. These two are exactly that.

Both are live in the current release — download s1ft or update your copy, and they’ll be waiting. If you’re new here, start with the basics and the rest takes care of itself.