Help
How PAuL works, feature by feature. If the answer you need is not here, write to support@macpaul.app — a question that reaches us twice usually means this page is missing something.
Getting started
What do I need to run PAuL?
Mac with Apple silicon (M1 or later), macOS 14 Sonoma or later. Transcription runs on the Apple silicon neural engine, which is why an Intel Mac is not supported.
Which permissions does it ask for, and why?
Microphone is needed to record anything at all. Screen Recording is needed for two separate things: capturing the other side of a call while you are on headphones, and reading a meeting window's title so a recording can be named after the meeting rather than the time of day. If you decline it, PAuL still records your microphone.
macOS only asks once. If you said no and changed your mind, the switches are in System Settings under Privacy & Security.
Where do the recordings go?
Your Downloads folder by default, and you can point PAuL somewhere else.
Each recording is an ordinary .m4a file with its transcript and
summary beside it as plain JSON. Nothing is in a database and nothing is in a
format only PAuL can read: if you stop using it, the recordings are still
yours and still open in anything.
Recording
Can it start on its own when a meeting begins?
Yes. Turn on Automatically record meeting calls in Settings. Recording starts when a meeting app opens your microphone and stops 45 seconds after the call ends. It works with the Teams, Zoom and Webex apps, and with Teams, Zoom, Webex and Google Meet in a browser — the browser case needs Screen Recording permission so PAuL can see the meeting window's title.
What does PAuL record: my microphone, or the call?
It depends on where the sound is playing, and PAuL works that out for you. On headphones or earbuds your microphone cannot hear the other people, so PAuL records the call's audio from the Mac directly, along with your microphone. On speakers the microphone hears the call out loud along with you, so PAuL records the microphone alone; recording the call directly as well would only add an echo of it.
It follows you during a recording too. Put earbuds in part-way through a call and PAuL switches over; take them out and it switches back. The changeover takes about a second.
The recording bar shows what is being captured. Incoming is the call, Outgoing is you. When there is a single meter labelled Microphone, the microphone is recording both.
Some devices cannot be told apart: an audio interface or a dock looks the same with headphones plugged into it as with loudspeakers. If PAuL reads yours wrongly, switch with the audio source button on the recording bar while recording. PAuL remembers that for the device and does not need telling again.
Does everyone on the call need to agree?
In many places, yes, and in some it is the law rather than a courtesy. PAuL does not announce itself to the other participants, so telling them is yours to do.
What is screen capture for?
With Capture screen during recording on, PAuL saves periodic screenshots of the meeting window, so a transcript that says "as you can see on this slide" still means something afterwards. They appear on a Screenshots tab beside the transcript.
Is there a limit on how long I can record?
No, and there is no per-minute charge, because the transcription happens on your Mac rather than on somebody's server. The only limit is disk space.
Transcription
Where does transcription happen?
On your Mac, using WhisperKit on the neural engine. The audio does not leave the machine, there is no account, and it works with the network off.
How long does it take?
Less than the recording itself on current hardware, and it runs in the background — you can carry on using PAuL while it works. A long recording queued behind others will take its turn.
Can I run it again?
Yes. Re-transcribe in the toolbar discards the existing transcript and starts over. The audio is never touched.
Speakers
How does it know who said what?
In two steps. First the audio is split by voice, on your Mac, which gives you Speaker 1, Speaker 2 and so on. Then, if you have set up a model under Ask, PAuL reads the transcript and works out which speaker is which person — that is the Labeled Transcript on the Speakers tab.
Why does it still say "Speaker 1"?
Because nobody said a name. The labelling works from what the call contains: if people greet each other by name, or introduce themselves, PAuL can attach names to voices. In a call where nobody is ever addressed by name there is nothing to go on, and PAuL says so rather than guessing. Use Identify Speakers in the toolbar to run it again.
What does "Split into Calls" do?
PAuL records a session, not a call, so one file often holds a call, a long pause, and then another call. Split into Calls finds those gaps and shows you each one with the words either side, to confirm or dismiss before anything happens. Each call you keep becomes a recording of its own with its share of the transcript and screenshots. The original is always kept, and the audio is copied rather than re-encoded, so the parts sound exactly like the original.
Summaries
What is on the Summary tab?
One line saying what the call was, a short paragraph you could read instead of it, the handful of things that were actually said, and anything anyone agreed to do. It is written once when the recording finishes transcribing and kept beside it, rather than worked out again every time you open it.
Why are there no action items?
Because the call did not produce any. PAuL reports what the call says: a conversation where nobody agreed to do anything is shown with nothing in that section rather than having something invented for it.
Can I summarise older recordings?
Yes — there is a button on each one. Nothing is run across your whole library unasked, because each summary costs you a call to your model provider.
Ask
What is Ask?
A question box that reads your transcripts and answers from them, with the recording and the timestamp each claim came from. "What did we decide about the migration?" is the kind of thing it is for. You can point it at one recording or at all of them.
Why does Ask need an API key?
Recording and transcription run on your Mac. Answering a question in English about forty hours of meetings does not: that needs a large language model, and running one locally would mean a far bigger download and a much slower answer. So Ask uses a model you choose, with your own key, and it is off until you set one up.
Which providers work?
Anthropic, OpenAI and AWS Bedrock. Anthropic and OpenAI need an API key, which you paste into Settings once. Bedrock uses an AWS profile already configured on your Mac.
Where does my key go?
Into the macOS Keychain, the same place Safari keeps passwords. It is not in a settings file, not in the app bundle, and never sent anywhere except the provider it belongs to.
What does it cost to run?
Whatever your provider charges for the tokens, billed by them to you. PAuL adds nothing. A question about one recording is a fraction of a cent; a question across a large library costs more, because more transcript is read.
What leaves your Mac
Plainly, what is sent anywhere?
Recording, transcription and the speaker split happen entirely on your Mac. No account, no sign-up, no telemetry, nothing in the background.
Three things do use the network, all of them things you turn on:
- Ask, summaries and speaker naming send transcript text to the model provider you chose, using your key. That is the feature — it cannot answer questions about your meetings without reading them. Audio is never sent, only text.
- Checking for updates asks macpaul.app whether a newer version exists.
- A crash report, if one happens, is shown to you first and sent only if you press the button.
If you never set up a model, PAuL records, transcribes and separates speakers with the network off.
Does my provider train on my meetings?
That is between you and them, and their policy rather than ours. API access at Anthropic and OpenAI is not used for training by default; Bedrock runs in your own AWS account. Check the terms of the one you pick if it matters.
Trial and licence
How does the trial work?
14 days, everything working, no card. After that, recording stops until you buy a licence; the recordings you already made stay where they are and remain readable.
What does it cost?
$39 once. Not a subscription, and transcription has no per-minute fee, so a heavy month costs the same as a quiet one.
I have a new Mac. Can I move my licence?
Yes. Install PAuL and paste the same key. If you have lost it, write to support@macpaul.app from the address you bought with.
When something goes wrong
macOS says it cannot verify the app
That means the download did not finish, or it came from somewhere other than macpaul.app. Every release is signed with an Apple Developer ID and notarised by Apple, so a complete download from this site opens normally. Download it again from macpaul.app.
There are no screenshots in my recording
Either Capture screen during recording is off, or macOS has not been given Screen Recording permission. Both are worth checking before the next meeting rather than after it.
The transcript only has my side of the call
You were on headphones and PAuL was recording only your microphone. From version 1.0.68 PAuL captures the call itself whenever sound is on headphones, so the usual cause now is that macOS has not given PAuL Screen Recording permission, which is what recording the Mac's own audio needs: allow it under System Settings › Privacy & Security › Screen & System Audio Recording. Existing recordings cannot be recovered — the other side was never captured.
Ask says it needs a provider
Open Settings, choose the LLM tab, pick a provider and paste a key. The status line reads Ready when it is set up.
Still stuck
Write to support@macpaul.app and say what you did, what happened, and the version from the bottom of the sidebar. If PAuL crashed, send the report it offered — it says what failed, and it is the difference between a fix and a guess.