Two jobs, in detail
The short version is on the home page. This is the whole of it — how dictation behaves, what a meeting leaves on disk, and how long it takes.
Hold a hotkey, speak, release.
Hold the key and speak, or switch to tap-to-start; the key is yours to choose. Your words are on the clipboard about a second after you finish. Switch on Insert at cursor and Saykeep pastes them (⌘V) into the field you were typing in — any Mac app that takes a paste: your editor, your mail, your terminal, your chat. That is what the Accessibility permission is for. On the clipboard path, clipboard-history tools will see your dictation like any other copy.
A custom vocabulary — a plain text file of your names and jargon, kept on your Mac — fixes their spelling in dictated text and in what the summariser reads, and hands Whisper the same list as a glossary hint while it transcribes. The saved meeting transcript stays what the model actually heard.
The on-device decode of a dictation of up to 30 seconds, once the model is warm: median 0.89 s across 101 dictations from the developer's own logs on one Apple M5 Mac (32 GB), Whisper large-v3-turbo on the GPU — fastest 0.6 s, slowest 10.5 s, measured before the text is pasted. Longer dictations take proportionally longer; there is no cap. On a memory-pressured Mac the first dictation after a long idle has taken several seconds longer while macOS paged the model back in; since 0.3.7 that reload starts the moment you press the key (not yet re-measured).
The default models on a 100-sentence sample of the public FLEURS test set per language — read speech, not meeting audio — transcribed on an Apple Silicon Mac with stock settings: roughly one word in twenty wrong, in both languages. Names and specialist terms are where any model slips; that is what the custom vocabulary is for.
Whisper large-v3-turbo for dictation, Whisper large-v3 for meetings, both running on your Mac's GPU through mlx-whisper. Dictation is the model's own output plus your vocabulary fixes — punctuated, but no rewrite pass and no cloud. Dozens of languages, auto-detected per recording or pinned to one in Settings; pick a smaller model there if you want speed over accuracy.
No bot in the call. No cloud behind it.
Start a recording and Saykeep captures your mic and everything your Mac plays, straight from macOS's own audio system — no bot joins the call, no virtual audio driver to install. The capture is system-wide, so it doesn't care which app is talking: Zoom, Meet, Teams, a Slack huddle, a browser tab. Zoom is what the developer records with every day. For meetings, macOS asks for two permissions: Microphone and System Audio Recording.
When the call ends, Saykeep keeps working in the background
On the Mac we measured (an M5), an hour of meeting audio works out to roughly 3–7 minutes of transcription and about 30 more of speaker labelling, which runs on the CPU before the transcript is written — scaled from real 10–17-minute recordings. Older Apple Silicon Macs will be slower; we haven't measured them yet.
A summary has three parts
Topics · Decisions, each marked Decided or Discussed · Action items, where the model is instructed to tag an owner on every item: the person at the microphone, a participant named out loud, or Unassigned. The text is written by the AI endpoint you connect; without one, the summary step is skipped.
## Topics - … ## Decisions - [Decided]: … - [Discussed]: … ## Action items - [Marek]: … - [Unassigned]: …
Speaker labels, computed on your Mac
One ~30 MB model download the first time, then fully offline. Open a recording, rename “Speaker 1” to a real name, and that meeting's transcript re-renders with it. Voices picked up by your own mic are marked “You” unless you turn on mic speaker identification in Settings.
Recording is visible by design
Whenever Saykeep is recording a meeting, the menu-bar icon shows it the whole time, and no setting turns that off — a test sweeps every setting to keep it so. Before your first meeting recording, a plain-language notice says recording may need every participant's consent and names jurisdictions to be aware of. The indicator is on your screen only: Saykeep sends nothing into the call, so telling the other people is still up to you. Is it legal to record?
Audio you can let go of
Keep the audio (the default), have Saykeep delete it once the meeting's transcript is written, or have it ask you after each meeting. Audio is only ever deleted after a transcript exists; transcripts and summaries are never deleted automatically.