Guide

Transcribe and Summarize Recordings on iPhone and Mac — Without the Cloud

You have a recording — a meeting, a lecture, an interview, a voice memo, maybe a video — and you want it as text you can search, quote, and ask questions about. Most tools that offer this route your audio through someone else's servers. This guide covers the alternative: transcription that runs entirely on your iPhone, iPad, or Mac, what Apple already gives you for free, and what to do once the transcript exists.

Why recordings deserve the off-cloud default

A recording is different from a document you wrote. It contains other people's voices — things they said in a specific room, to a specific audience, with specific expectations. A weekly standup covered by an NDA. A user-research interview conducted under a consent form that named who would hear it. A client call. A lecture where classmates asked questions they did not expect to be uploaded anywhere.

When you send that file to a cloud transcription service, you are making a data-handling decision on behalf of everyone in the recording. Sometimes that is fine — a public podcast has no confidentiality to protect. But for meetings, interviews, and anything work- or study-related, the safe default is the opposite: the audio stays on the device that holds it, and the transcription comes to the audio rather than the audio going to the transcription.

There is also a mundane practical upside: no upload step means large files (a two-hour lecture video, say) transcribe without a network connection at all — on a plane, in a basement lecture hall, on hospital Wi-Fi you do not trust.

What your iPhone and Mac already do — use it if it is enough

Before installing anything, know what Apple ships natively, because for many people it fully answers the question:

If your need is "I recorded a memo and want its text," stop here — Voice Memos is the answer, it is free, and it is already on your phone. This page will not pretend otherwise.

The native tools stop at three specific walls:

  1. They transcribe what they record. A file that already exists — an exported Zoom recording, a lecture video, an MP3 a colleague sent, years of old interviews — has no clean path into Voice Memos or Notes for transcription.
  2. Each transcript lives inside its one recording. You can find a phrase, but you cannot ask a question — there is no way to ask something across twenty recordings and get an answer that tells you which recording, and where, it came from.
  3. No citations, no restraint. A native summary gives you a paragraph; it does not point at the exact passage it came from, and it will not tell you when the recording simply does not contain what you are looking for.

Can ChatGPT summarize an audio recording? The honest answer

Yes. Cloud AI assistants can transcribe and summarize audio, and generally do it well. If capability were the only question, this page would be shorter.

The real question is routing. To summarize your recording, a cloud service has to receive it: the file is uploaded and processed on the provider's servers, under whatever retention, logging, and training policies apply at that moment — policies you would need to read, and re-check, because they change. None of that requires assuming any provider is careless. The point is simpler: the confidentiality decision is made the instant the file leaves your device, before any policy on the other end matters. For an NDA-covered meeting or a recorded patient conversation, the upload itself can be the violation, regardless of what happens to the file afterward.

So the honest framing is: use cloud AI for audio that has no confidentiality to protect, and use on-device transcription for audio that does. (If you want to know how to check what any AI app actually collects — rather than taking its marketing's word for it — see what "Data Not Collected" actually means.)

Transcribing an existing audio or video file on-device, step by step

The gap the native tools leave — import a file you already have, transcribe it locally, then actually work with the text — is what OpenIntelligence is built for. It is free to download on the App Store, with one requirement stated plainly up front: it needs iOS, iPadOS, or macOS 26 or later on Apple Intelligence-capable hardware — an iPhone 15 Pro or newer, or an iPad or Mac with an M1 chip or newer. On older devices it will not run, and the native options above are the honest recommendation instead.

  1. Install the app and create a collection. A collection is just a bucket — "CS 301 Fall," "Acme project," "Interviews."
  2. Add the recording. From the Files app or the share sheet on iPhone and iPad, or by dragging files in on the Mac. Audio and video both work — for a video, the soundtrack is what gets transcribed.
  3. Let it transcribe locally. Speech recognition runs on the device using Apple's on-device transcription. Nothing is uploaded — there is no account and no API key to even send anything to. If you want proof rather than a promise, turn on airplane mode first; the transcription proceeds anyway. How long it takes depends on the recording's length and your hardware.
  4. The transcript is indexed alongside your documents. The text goes into the same local index as any PDFs, slides, or notes in the collection — full-text search via SQLite FTS5 plus semantic retrieval, accelerated across the CPU, GPU, and Neural Engine.
  5. Ask questions. Answers come with tappable citations pointing at the exact transcript passage they came from. And when your recordings do not contain the answer, the app abstains and says so — it does not fill the gap with a guess. If you just want the raw transcript, that is there too.

After the transcript: search a semester, not a file

A transcript on its own is a wall of text. The useful version of this workflow starts when recordings and documents share one index.

The clearest case is a student's: put twelve weeks of lecture recordings and the professor's slide decks into one collection. Now "when did we cover eigenvalues, and what did the slides add that the lecture skipped?" is a question you can actually ask — and the answer cites the specific lecture's transcript segment and the specific slide page, so you can jump to either and verify. Scanned handouts work too; printed pages are read with on-device OCR before indexing.

The same shape works for meetings: the recording of the negotiation next to the contract being negotiated, or a quarter of project calls next to the spec they kept referencing. The mechanics of asking questions across many documents at once — and how the citations and abstention behave — are covered in searching inside multiple PDFs at once, and the broader picture of what local models can and cannot do is in what on-device AI can actually do.

What this is not

To save you a download if you are looking for something else:

One more honest limit that applies to every transcriber, local or cloud: quality tracks the audio. A phone in the middle of a conference table, heavy crosstalk, or a bad microphone will degrade any transcript, and Apple's on-device models are no exception. Skim the transcript before you quote someone from it.

Requirements and limits

If the native transcript in Voice Memos covers your need, use it. If your recordings are files you already have, or the transcript is the beginning of the work rather than the end of it, that is the case this workflow — and this app — exists for.

Related reading

On-Device AI on iPhone, iPad, and MacWhat local models can actually do today — and where the honest limits are. Search Inside Multiple PDFs at OnceAsk one question across a whole folder of documents and get cited answers. What "Data Not Collected" Actually MeansHow to verify any AI app's privacy label instead of trusting the marketing. OpenIntelligenceOn-device document AI for iPhone, iPad, and Mac — answers with citations, offline.