The room and the call, separately
On macOS the mic and the system audio are two tracks, segmented and transcribed independently, then merged by timestamp. Overlapping speech stays readable instead of turning into one scrambled line.
macOS · local-first · one-time purchase
Coii Audio records the room and the call, transcribes both as they happen, and writes the summary — on your machine. No cloud, no account, no subscription.
30 days free, no card · then $19 USD once · macOS 13+, Windows, Linux
Weekly product sync
Today · 42 min · 3 speakers
REC 42:11Setup screen — the download step
Your notes
☑ ask design for the three-row layout
☐ check the disabled state copy
Nothing is uploaded
No server of ours
No account
Nothing to sign up for
Runs offline
After the first download
$19 USD once
Updates included, forever
How it works
Capture, transcription and the language model all live in the same application. There is no sidecar to install, no Python, no Ollama and no API key.
Your microphone and the system audio are captured as two separate tracks — the room and the call, kept apart.
Whisper runs on your GPU, segment by segment, while the meeting is still happening. Voice prints tell the speakers apart.
A language model inside the app turns the transcript — and your notes — into minutes you can question afterwards.
Features
Two people on a call, four people around one laptop, or a three-hour workshop in a second language — the awkward cases are the ones that shaped this.
On macOS the mic and the system audio are two tracks, segmented and transcribed independently, then merged by timestamp. Overlapping speech stays readable instead of turning into one scrambled line.
The track says which side of the call a segment came from. Voice prints say who, inside a track — clustering, not recognition, so names are yours to type and never leave the session.
Write during the meeting and the summariser organises the minutes around what you wrote, expanding your shorthand from the transcript rather than starting from scratch.
Questions are answered from the transcript, grounded in it, and the ask box is docked in the live view too — you can ask before the meeting is over.
A Chinese meeting gets a Chinese summary under Chinese headings. The language is decided from the transcript, not from the app's own locale.
Long transcripts are summarised in chunks and then synthesised once, so a long meeting doesn't need a long context window — or your patience.
What it sends
“Private” is easy to write on a page, so here is the complete list instead. No analytics, no crash reporting, no usage statistics, no update check, and no connection at launch.
01
When you download a model
huggingface.co and github.com see your IP and which model file you asked for. Once it is on disk they are never contacted again.
02
When you press Activate
Lemon Squeezy receives your licence key, and a device label when you activate. At most three times per device — check, activate, deactivate.
03
While the trial is running
A HEAD request to a large CDN, read for its Date header only, so a wound-back system clock can't buy free days. Offline is not an error.
04
Whenever the app thinks
The language model listens on 127.0.0.1. That traffic never reaches an interface that could carry it off the machine.
Audio is never written to disk — not even a temporary file. What remains of a meeting is its text, in a database you own.
Read the privacy policyPricing
Thirty days with every feature unlocked, no card and no account. Then $19 USD, once.
Personal licence
$19USD once
Not a subscription. Nothing renews.
You can buy from inside the app at any point in the trial.
Secure checkout via Lemon Squeezy, our merchant of record. Cards and more, tax handled, invoice included.
Lost your key? Lemon Squeezy’s My Orders mails it to you again — sign in with the address that bought, no email to us required.
Questions
Download it, record the next call, and read the minutes before anyone has asked for them.