Recording a meeting
into a note.
How to record, what the note contains, and every setting behind it — the speech model, the two modes, retention, the language override, and the one thing that can leave your Mac.
Four steps, and
only one is new
Downloading the speech model is a one-time thing. After that, recording is a slash command.
Download a speech model
Settings → Meetings → Speech model. The accurate one is about 570 MB, the fast one about 140 MB; both are downloaded once and kept on your Mac. If you start a recording without one, Ampersand tells you before the meeting rather than after it — and records anyway, so nothing is lost.
Start recording
Type /record in a note to drop a recorder card where your cursor is, or use the menu-bar icon or ⌥⌘R when no window is in front of you. Both ask which kind of recording it is.
Pick in person or a call
In person records the microphone only. Online call additionally captures the audio your Mac is playing, so the other side of a Zoom, Meet, Teams or FaceTime call is recorded too — that mode asks for macOS Screen Recording permission the first time.
Stop and let it write
Transcription has been running in the background the whole time, so the summary begins appearing within seconds and streams into the note line by line. One undo removes the whole thing if you want to start over.
On this Mac,
and then nowhere
Speech is transcribed by a Whisper model running on your own machine, on the GPU. The audio is written to a working folder outside your vault — so it is never carried into iCloud or whatever else syncs that folder — and, on a Mac that can store a key for it, encrypted at rest with a key belonging to that machine alone. By default it is deleted as soon as the transcript exists.
★ The one part that can leave your Mac is the transcript. ★ The summary is written by whichever AI provider you configured, so if that is Claude or OpenAI, the transcript is sent there with your own API key — and a transcript contains what other people said, who never configured anything. Settings → Meetings names the provider in plain text whenever the one you have chosen is off-device, and says nothing when it is local, because a warning on the safe route is how people learn to ignore the one that matters. Choose Apple Intelligence or a local model through Ollama and nothing leaves the machine at any point. See the privacy policy for the full picture.
Whether you may record a conversation, and whether you have to tell the other people first, depends on where you and they are — and it is your call, not the app's. Ampersand records the way a voice recorder on the desk would: you are the one operating it, nothing joins the call, and macOS shows its own orange microphone indicator the entire time.
What you can
change
All of it lives in Settings → Meetings, except the summary provider, which is part of the normal AI routing.
Speech model
Whisper large-v3-turbo is the default — accurate, and fast enough on Apple Silicon because it runs on the GPU. The base model is roughly four times smaller and noticeably rougher; it is there for older Macs and tight disks.
Summary provider
The summary is written by whichever AI you have configured, using the “Meeting summary” routing key in Settings → AI. Claude, OpenAI, a local Ollama model, Apple Intelligence and CLI agents all work. Turn summarizing off entirely and you still get the transcript.
Spoken language
Automatic detection is right most of the time. When it is wrong, Whisper does not fail — it TRANSLATES, silently, into fluent English. Setting the language explicitly is what makes that recoverable, so do it if you work in a language other than English.
Retention
Delete after transcription is the default. Keep the audio for 7 or 30 days if you want to re-listen to anything the transcript got wrong, or indefinitely — Settings lists every recording with its size and lets you delete any of them by hand.
Screen capture
Off by default, and only available on calls. When on, Ampersand samples occasional still frames of the shared screen, reads the text on them on-device, and puts them in the note. A screenshot of someone else’s screen often carries more than the audio does, which is why it is a separate choice.
macOS only
Permissions
Free vs Pro
Good to know
Where is the note saved?
In a Meetings folder in your vault, named for the date and time — both configurable in Settings → Meetings. If you started the recording with /record inside a note you were already writing, the transcript lands in that note instead, right where the card was.
What is actually in the note?
A short summary, the decisions that were made, the action items as real checkboxes, and the complete transcript folded into a collapsed toggle underneath. It is ordinary markdown — the toggle is a details element, the tasks are task-list items — so everything works with the rest of the editor, with export, and with any other markdown tool.
Why does the transcript not say who was speaking?
On a call it does: your microphone and the far end are separate audio tracks, so “you” and “them” is a fact rather than a guess. In a room everyone shares one microphone, and telling voices apart reliably needs speaker fingerprinting that would sometimes be confidently wrong. Rather than guess, Ampersand leaves an in-person transcript unlabeled and says so in the note.
Can I record while my vault is locked?
Yes. Recording and transcription never consult your vault passphrase — the audio is protected with a key belonging to this Mac, not with your vault key. Writing the finished note into an encrypted note still needs the vault unlocked, like any other write.
What if transcription or the summary fails?
The recording is kept. Audio is only ever deleted once its content exists somewhere else, so a model that was unreachable or a crash between stopping and writing cannot cost you the meeting. Settings → Meetings lists recordings that never produced a note, so you can see what happened rather than discovering it later.
The transcript got a number wrong.
It happens — speech recognition turns “fifty” into “fifteen” sometimes, and no model is exempt. This is the real cost of the default retention setting: with the audio deleted, the transcript is the only record and a misheard figure is unverifiable. If numbers and names matter in your meetings, set retention to keep the audio for a week.
Does it need the internet?
Not for recording or transcribing — both are entirely local. Only the AI summary needs a connection, and only if you chose a cloud provider for it.
Related guides: the summary is written by whichever AI provider you have chosen, which is also what decides whether the transcript stays on your Mac. The note itself is an ordinary Markdown file you can export or publish, and it lives in the vault folder like any other note.
Be in the meeting,
not in your notes.
Download Ampersand, download a speech model once, and record your next conversation into your vault.
Free to start · Mac, Windows & Linux · No account required