Frequently asked questions

Which Macs does it run on?

A Mac with Apple Silicon (M1 or later) and macOS 14.2 or later. Intel Macs are not supported. Macs with 8 GB of memory: not measured yet, so try it free first. Saykeep for Windows is in testing, with no download, date or price yet (the Windows page). Only have a Windows PC? Don’t buy yet.

How much disk space does it need?

About 4.7 GB of speech models, plus a ~30 MB speaker-label model, downloaded once on first run behind a progress window (it counts in binary units, so its two counters add up to about 4.4 GB). The app download itself is 329 MB. Have about 5 GB free. A model already on your Mac is not downloaded again, and saykeep.app/network lists where each download comes from.

What do I need for summaries? Does my ChatGPT Plus work?

Summaries need a model you connect: a free one on your Mac (LM Studio or Ollama), or your own API key for an OpenAI-compatible service, which then receives the transcript text (plus your name and the optional “about you” note, if you set them). ChatGPT Plus doesn’t include API access, so it can’t be used here. Without a model you still get the full transcript with speaker labels; only the summary is skipped. No endpoint is set until you choose one, and Saykeep never bundles or downloads a language model. We also plan an optional Saykeep-provided endpoint for people who’d rather not run or rent one themselves. It would be exactly that, optional: the app you buy doesn’t need it.

Which languages does it transcribe, and how accurately?

We measured two: English and Polish, each at roughly 5 % word error rate (about one word in twenty wrong) on a 100-sentence sample of the public FLEURS test set per language, on one M5 Mac. That is read speech, so real meetings score worse. Polish and English in the same call: not checked yet. Settings ▸ Transcription ▸ Spoken language lists other languages too; we haven’t measured them, so we make no claim about them, and the 30-day trial is the way to check yours.

Names and jargon are where every model slips. That is what custom vocabulary is for, and transcripts are plain text you can edit. Speaker labels are good but not flawless, so you can rename and fix them. Settings also offers a faster engine, Parakeet: its maker lists 25 European languages (their figure, not our measurement), and it detects the language itself, can’t be pinned to one and ignores your custom vocabulary.

How long after a call is the transcript ready?

About 35 minutes per hour of meeting with speaker labels on, which is the default: labelling runs on the processor before the transcript is written. With speaker labels switched off (Settings ▸ Transcription ▸ Speaker labels), about 3–7 minutes per hour. Both were measured on an M5 and scaled from real 10–17-minute recordings. Older Macs are slower; we haven’t measured them yet. It all happens in the background after the call.

Is Saykeep finished?

Pre-1.0, and we say so. The current release is 0.4.3, and the changelog lists what changed in each one. 0.x releases count as v1 for your licence (the app’s licence major is 1), so the key you buy unlocks them just as it unlocks every v1.x release. A future v2 may be an optional paid upgrade at an owner discount.

What can connect to the internet?

Speech-to-text, transcripts and speaker labels are made on your Mac. There are no accounts, and the app sends us nothing about you. The complete list of what can connect is short and published on saykeep.app/network, with when each connection happens and what it sends. In practice:

• 1. Model downloads. The speech models download once from Hugging Face and its download servers, and the ~30 MB speaker-label model once from models.saykeep.app, with no account. A model already on your Mac loads from disk with no request (0.4.0 sent one small metadata request on every load; 0.4.1 removed it).
• 2. Your summary endpoint, only once you set one. When a summary is made, it receives the meeting’s transcript text, plus your name and “about you” note if you filled them in. Each time you open the menu-bar panel or press Test, it also gets a model-list request, carrying your API key if you stored one. Point it at a server on your own Mac (127.0.0.1) and the text goes only there.
• 3. The update check, if you said yes at setup (Yes is pre-selected). At most one request a day to saykeep.app, carrying your Saykeep version and nothing else. An update downloads from dl.saykeep.app only when you click Install Update.

The source document, including the optional Hugging Face token fallback, is reproduced on saykeep.app/network. You don’t have to trust us: run Little Snitch or LuLu and compare what you see with the list. A connection that isn’t on it is a bug, so tell us. This website is separate from the app: the privacy page says what it loads.

Does Saykeep send telemetry or analytics?

Saykeep sends no telemetry, analytics or crash reports, and has none of its own. On the Mac, some of the open-source libraries it bundles can report usage. The speaker-label library (pyannote) has it on by default, and Saykeep switches it off before loading it. The model-download library (huggingface_hub) has a usage flag Saykeep does not yet switch off: nothing in Saykeep calls its reporting helper, but the model-download requests it makes to huggingface.co can carry version numbers in a header (its own, Python’s and torch’s). saykeep.app/network lists each library and what the version you download does about it.

Where are my recordings saved? Do they go to iCloud?

In ~/Documents/Saykeep, as ordinary files: a folder per meeting with the audio (Opus), the transcript (JSON and Markdown) and the summary. If your Mac syncs Desktop & Documents to iCloud, that folder syncs too, to your own iCloud account; we never receive it. To keep the files on your Mac only, switch that sync off: in System Settings, under your name ▸ iCloud, turn off Desktop & Documents Folders for iCloud Drive. macOS keeps the files it already synced in iCloud Drive, so move them back into Documents on your Mac.

What is kept, and can Saykeep delete the audio for me?

Settings ▸ Privacy & Storage ▸ Recordings decides what happens to a meeting’s audio once it is transcribed: keep it (the default), delete it once the transcript is written, or have Saykeep ask after every meeting. Audio is only deleted after a transcript exists, so a failed transcription never costs you the recording; transcripts and summaries are kept whichever you pick. You can also delete any meeting from the Recordings window, or erase everything (every recording, transcript and summary, plus Saykeep’s settings, models and stored keys) in one step that asks you to type ERASE.

Why does the icon show a warning when my model server is off?

Each time you open the menu-bar panel, Saykeep checks whether the summary endpoint you set is answering. If your model server is off, the icon turns amber and the panel shows a banner saying the endpoint can’t be reached. Nothing is lost: meetings still record and transcribe, and a summary that couldn’t be made is kept in the queue and retried. Once the server is back, press Test in Settings ▸ Summaries & AI and it is made straight away. No endpoint set? Then there is nothing to check, and no warning.

Why does macOS show a recording indicator between meetings?

After your first meeting, macOS’s system-audio recording indicator stays lit until you quit Saykeep. Saykeep keeps its connection to the Mac’s audio open between meetings, because rebuilding it for every meeting is what knocked macOS audio over. Nothing is saved or sent between meetings: nothing is written until you start the next one. Quit Saykeep and the indicator goes off.

Is it legal to record my meetings with Saykeep?

It depends on where you and the other participants are — and it’s your responsibility, not ours. Some places are all-party-consent jurisdictions, where every person in a private conversation must agree before it’s recorded. Germany is one: recording someone’s non-public spoken word without consent is a criminal offence there (§ 201 StGB). Several US states are too, California among them. Most other US states and US federal law follow a one-party rule, and when participants are in different places the stricter rule can apply. Check the law where you are. This FAQ is not legal advice.

What Saykeep does is make recording unmistakably visible so consent is possible in the first place: the recording indicator is always on, and your first meeting shows a consent notice naming the jurisdictions to be aware of. Obtaining the actual consent is up to you.

Can other people in the meeting tell I’m recording?

Not automatically — Saykeep sends nothing into the call, so it can’t announce itself to the other side. What it refuses to do is hide the recording from you: there is no hidden or low-visibility recording mode, and no setting turns the indicator off. Whenever meeting capture is active, the recording state is shown on your screen: the menu-bar icon, the status line (elapsed time plus live audio meters), and the recording card. That’s deliberate. The indicator means you can’t record by accident. Announcing the recording to the other participants is still your job, and in all-party-consent places, your legal obligation.

How much does it cost?

$39, one time. One payment, no subscription, every feature included. There’s a 30-day free trial (all features, no card, no account) and a 30-day money-back guarantee if you buy and change your mind. The checkout shows your total, including any tax. See the price and buy.

What would two years of the alternatives cost, at list price?

Saykeep is $39 once. Two years at each vendor’s list price:

Not like-for-like: Wispr Flow, Otter and Granola transcribe in the cloud, and Saykeep’s summaries need a model you connect. List prices from each vendor’s page, checked 1 October 2026. Wispr Flow, Otter and Granola also have free plans. Why pay when free tools exist?

When do I get my licence key?

Your key comes in a separate email from Piotr Januszek, not in the Lemon Squeezy receipt — by hand, usually within a day; there’s no activation server, which is also why the key is checked on your Mac. Use the trial meanwhile. If the trial has already ended: Settings ▸ App & Updates ▸ Licence ▸ “I’ve paid — waiting for my key”, with the order number from your Lemon Squeezy receipt, gives 7 days of full access, once per Mac. Nothing after 24 hours? Check spam, then email support@saykeep.app with your order number. Don’t buy again.

How do I activate it? Do I need to be online?

No network needed. Paste the key in Settings ▸ App & Updates ▸ Licence and press Activate. It is checked on your Mac against a public key built into the app, with no network call: there is no activation server. You can activate on a plane.

What happens when the trial ends?

The trial runs 30 days from first run. When it ends, new capture (dictation and meeting recording) is gated until you enter a licence key, but you can still open and browse everything you already recorded. We gate compute, never your files. Bought, and still waiting for your key? Here’s what to do.

New Mac or a reinstall: do I lose my licence or restart my trial?

Your key activates again on any Mac you personally use. Paste the same key into Settings ▸ App & Updates ▸ Licence on the new install or the new Mac and you’re licensed again. The check happens on your Mac, so the key is all you need.

The trial clock is stored in your Mac’s Keychain, so it survives a reinstall: reinstalling does not grant a fresh 30-day trial. This is deliberate.

What do future updates cost?

Every v1.x update is free to you as an owner, through at least the current and the next major macOS release. A future v2.0, reserved for a genuinely large new local capability, may be an optional paid upgrade at an owner discount. Your v1 licence never expires; it just doesn’t unlock v2.

Technically, your key carries a major-version entitlement: a v1 key validates every v1 release and does not unlock v2. Updates install in-app, only with your consent, and there is no subscription.

Why does my card statement say “LEMSQZY*”?

Checkout is processed by Lemon Squeezy, our merchant of record: they take the card payment and handle VAT, so your statement shows their descriptor, not ours. Your statement will read LEMSQZY*. We never see your payment details. Your 30-day money-back guarantee applies regardless. For a refund, email support@saykeep.app.

Can I get an invoice for my company?

Lemon Squeezy is the seller and issues the invoice. Not sure it will suit your company’s books? Ask support@saykeep.app before you buy.

What happens if Saykeep shuts down?

The copy you’ve installed needs no server of ours: your key is checked on your Mac, and your files are open formats. Recordings, transcripts and summaries are ordinary files in your own folder (Opus audio, JSON and Markdown transcripts) that you can read without us. There is no subscription to cancel.

Why pay when free tools exist?

Only dictate? Handy is free and good. MacParakeet, free and GPL, labels up to eight speakers after a call too. Saykeep is for client calls you’d rather not send to a notetaker’s cloud, and it puts in one app what you would otherwise assemble: your mic and what your Mac plays, captured straight from macOS with no bot in the call; a transcript with who said what, where you name the speakers afterwards; dictation on a held key; summaries through a model you connect, and the transcript is there even when that model is off; an app signed and notarized by Apple; a human at support@ and a 30-day money-back guarantee.

My Mac’s audio dropped during a call. What do I do?

A known macOS issue: if another call starts or ends while Saykeep is recording, macOS audio can drop for a few minutes, in every app. Stop the recording before you hang up. On AirPods or Bluetooth days, choose your Mac’s built-in mic in Saykeep; that makes the drop less likely. If it happens, Saykeep tells you, keeps recording, and picks the audio back up when macOS recovers (up to ten minutes in our tests).

You don’t have to wait. When Saykeep detects the condition, it shows you a one-line fix that restarts the Mac’s audio service: open Terminal, type sudo killall coreaudiod, press Return and enter your Mac password, which won’t appear as you type. Audio comes back in seconds, in every app. Restarting Saykeep usually does not clear it.

Virtual or aggregate audio devices and pro audio interfaces: not tested. The trial is the test.

Is there a Windows version?

Saykeep for Windows is in testing. It has been built and run on real Windows 11 hardware, but testing isn’t finished, so there is no download, date or price yet. Only have a Windows PC? Don’t buy yet: join the list and get one email when it ships.

Something isn’t working. What should I send?

Export a diagnostics file: in Settings ▸ App & Updates ▸ Diagnostics, click Export diagnostics…; the file is written on your Mac, with no network call. Look it over, then email it to support@saykeep.app with what you did and what happened. The support page says what else helps.