Local AI note takers for Mac: what stays on it

Published September 28, 2026 · The Memory Notch team

Which Mac note takers transcribe on the Mac?

A local AI note taker for Mac turns speech into text on your own computer, so the audio doesn’t go to a transcription server. On September 23, 2026, Goodnotes’ parent company launched Goodmeet, a free beta that does this on Apple silicon. Four other Mac apps already did, and the five split on one thing: where the summary gets written.

• Summaries written locally: Meetily, Anarlog and MacWhisper, when you pick a local model.
• Summaries written elsewhere: Goodmeet uses a sealed cloud environment it calls confidential computing.
• No summaries at all: Memory Notch, our app. It records and transcribes, and stops there.

Granola is not on the list: it records on your computer but sends the audio to cloud transcription providers.

Checked September 28, 2026. Every price below comes from the vendor’s own page.

What counts as a local note taker?

A note taker is local when the speech-to-text step runs on your own computer. Recording the audio there is a separate step. An app can record on your laptop and still upload the sound for transcription a second later.

So every meeting app answers three separate questions, and a vendor can say yes to one and no to another:

1. Capture. Does a bot join the call, or does the app record your microphone and system audio? Our post on bot-free AI note takers covers this step.
2. Transcription. Does the speech model run on your computer or on a server? This post is about this step.
3. Summaries and chat. Does the model that writes the summary run locally, on the vendor’s servers, or with a provider you picked?

Granola shows why the split matters. Its security page says it “doesn’t store the audio from meetings” and uses “transcription providers (like Deepgram and Assembly)”.

So there’s no bot and no stored audio, but the sound still leaves your laptop to be transcribed. That’s a reasonable design, just not local transcription. Our Granola alternatives post goes further.

Which Mac apps transcribe on-device?

Five apps answer yes to the transcription question today. The table sums up what each vendor states on its own page. If you want to know how the speech models behind them compare, see Parakeet vs Whisper.

Goodmeet

Goodmeet says transcription “is done completely on-device”, that it doesn’t store meeting audio, and that “nothing extra joins the meeting”. It works with any call app.

Goodmeet comes from the company behind Goodnotes. It’s free during the beta and English-only for now, according to launch coverage. What to check: summaries are generated off your computer, and there’s no published price for after the beta.

MacWhisper

MacWhisper transcribes files and meetings with local models. It can start recording automatically when a call begins, and it labels speakers.

You can also connect a cloud AI provider with your own key, or run a local model. Pro costs €64 once and includes lifetime updates. What to check: which provider each feature uses once you add API keys. See our MacWhisper comparison.

Meetily

Meetily is a free, open-source meeting assistant for macOS and Windows. It transcribes with Whisper or Parakeet models and recommends Ollama for local summaries, with cloud providers as an option.

The paid Pro plan costs $10 per user per month, billed at $120 a year, and speaker labels are listed as a Pro feature. What to check: which summary provider is selected. See our Meetily comparison.

Anarlog (formerly Hyprnote)

Anarlog is an open-source meeting notepad that keeps notes and transcripts in a local database. The free plan includes “unlimited on-device transcription”.

Its AI features can run locally, through your own API keys, or through its hosted AI service. Pro costs $15 a month, or $150 a year, and adds cloud transcription. What to check: both the transcription setting and the AI setting. See our Anarlog comparison.

Memory Notch

Our app records your microphone and system audio from the notch and writes a live transcript with numbered speaker labels. The speech model is NVIDIA’s Nemotron multilingual streaming model, and speaker labels come from NVIDIA’s Nemotron 3 Diarization.

The two NVIDIA models run locally through an open-source toolkit called FluidAudio, which ships inside the app. We write no summaries, and each recording is saved as plain files in ~/memory.

What doesn’t “local” cover?

“Local” usually describes transcription only. Summaries, chat and optional cloud models are separate, and each one can send your transcript somewhere else.

Summaries. Goodmeet writes summaries “in an isolated environment, using confidential computing”. Launch coverage says it’s sealed “so that no one, including us, can access” the content. That’s a strong protection, but it runs on Goodmeet’s servers, not your laptop. Meetily and Anarlog can keep summaries local, but only when a local model is selected.

Cloud models you add. Three of the five apps let you plug in OpenAI, Anthropic or other providers. Once you do, that provider’s data policy covers whatever you send it.

Updates and licences. Local apps still go online for other jobs. Ours makes three requests, and none carries a recording. It fetches a signed list of revoked licences. It checks for updates once a day, which you can turn off. And it downloads an update when you choose to install one. Our privacy policy lists them.

Consent. Local processing does not replace the consent of the people you record. Tell them, and get the permission your situation requires.

We write no summaries. If you want one, open the transcript file with Claude Code, Codex or Cursor, or paste it into a chat app. That step sends the text to the AI service you picked. The AI tools guide shows each workflow.

What Mac and macOS do you need?

Most of these apps are built for Apple silicon, and few publish a minimum macOS version.

Local transcription runs on your own chip, so an older machine with less memory can slow down during long calls. Our specs page gives our details. For any of these apps, the best test is a short recording on the laptop you’ll use for calls.

How can you check an app is actually local?

Turn off Wi-Fi, record a two-minute test call, and see whether the transcript still appears. If transcription stops without a connection, the app isn’t transcribing locally.

Then check four more things:

1. Read the privacy page for provider names. Deepgram, AssemblyAI, OpenAI or “our transcription partners” mean audio or text goes to a server.
2. Open the AI settings. Find which model writes summaries. A local transcript can still go to a cloud model at that step.
3. Find the files. Local apps keep transcripts in a folder or database on your disk. Ours go in ~/memory, one folder per recording.
4. Watch the network. A firewall that shows outgoing connections tells you which servers an app contacts while you record.

The offline transcription guide walks through the same checks for our app.

Where Memory Notch fits, and where it doesn’t

Our app suits you if you want local transcription, plain files you own, and no summary step. There’s no bot, and an Enhanced transcript option re-transcribes a saved recording more carefully.

The limits:

• There are no automatic summaries.
• Speakers are numbered, not named.
• Library search covers titles and metadata, not every word.
• Overlapping voices can be difficult to separate.
• System audio needs macOS Screen Recording permission.

If you want summaries written for you, Goodmeet or Meetily will do more. If you want the transcript to be the last thing any app makes, ours does that.

Memory Notch costs US$79 once, for up to five of your own Macs, with a year of updates. Versions released in that year stay unlimited after it, and there’s a 30-day refund. The free trial records up to five minutes at a time, with no account and no expiry.

How the five-minute trial works has the details, and why your meeting notes don’t need the cloud explains the reasoning.

Frequently asked questions

Sources

• Goodmeet: the AI meeting notetaker without privacy tradeoffs
• Finwire / Business Reporter: The AI notetaker that keeps your conversations private
• MacWhisper
• Meetily on GitHub
• Meetily pricing
• Anarlog on GitHub
• Anarlog pricing
• Security at Granola
• Memory Notch privacy policy
• Memory Notch specifications

Related posts

• Bot-free AI note takers: capture isn’t privacy
• Granola AI alternatives, by why you’d switch
• Parakeet vs Whisper, and why we use neither live
• Why your meeting notes don’t need the cloud
• Offline transcription on a Mac
• Compare Memory Notch with other tools

Download Memory Notch · More from the blog