If you want meeting notes without handing your conversations to a cloud service, the field on Mac is now genuinely good: Speechmark, Meetily, and MacWhisper process everything on-device, while popular tools like Granola, Fathom, Fireflies, and Otter are better than their reputation on some counts — and worse on others. Here's the honest map, graded on where your audio actually goes.
Full disclosure: Speechmark is our app. Every claim about another tool below is cited to that vendor's own pricing or privacy page and qualified "as of July 2026", because these things change. And because no single tool wins every situation, there's a section on when you should pick a cloud notetaker instead.
What actually counts as a "private" meeting notetaker?
"Private" gets used for three very different architectures, and the difference is where your audio is processed — not whether a bot appears in your call.
- On-device: recording, transcription, and summarization run on your Mac. No server ever holds your audio. Speechmark, Meetily, and MacWhisper work this way.
- Bot-free cloud: nothing joins the call, which is discreet — but the audio is still uploaded and processed on the vendor's servers. Granola and Fathom sit here.
- Cloud with a bot: a visible notetaker joins the meeting and everything lives in the vendor's cloud. Otter and (by default) Fireflies sit here.
This isn't a theoretical distinction. In August 2025 a federal class action was filed against Otter over its notetaker's recording practices — NPR's report describes the core allegation:
The suit accuses Otter.ai of "deceptively and surreptitiously" recording private conversations that the company then uses to train its transcription service, without permission from the people being recorded. — Class action complaint, N.D. California, as reported by NPR (August 2025)
Employers have noticed the same risk category: in Cisco's 2024 Data Privacy Benchmark study of 2,600 organizations, 27% had banned generative-AI tools outright at least temporarily, and 63% limited what data employees may enter into them. (Cisco's 2026 follow-up shows outright bans easing as governance matures — the data-entry limits are the durable part.) If your workplace has rules like these, an on-device tool is usually the only notetaker that clears them, because no meeting data is entered into any external service at all.
How the tools are graded
Four questions, in order of how much they matter for privacy: Where is audio processed? Who keeps the recording? Does the vendor train AI on your content? What does it cost, and forever or monthly? Features come after those — a searchable team archive is worthless if the architecture is disqualifying for your work.
The comparison, as of July 2026
| Tool | Processing | Audio kept by | Trains on your content? | Price (as of July 2026) |
|---|---|---|---|---|
| Speechmark | On-device (Mac) | You (original audio stored locally) | No — nothing leaves the Mac | $79 one-time, free tier |
| Meetily | On-device (Mac/Win/Linux) | You | No — local, open source | Free, MIT-licensed; PRO tier planned |
| MacWhisper | On-device (Mac) | You | No — local | €64 one-time Pro, free tier |
| Granola | Cloud (bot-free) | Granola's servers | Team-wide training opt-out listed on Enterprise ($35) | Free basic / $14 / $35 per user/mo |
| Fathom | Cloud (bot-free) | Fathom's servers | Yes by default, account-settings opt-out | Free tier; paid plans |
| Fireflies | Cloud (bot by default) | Fireflies' servers | No, unconditionally | Free tier; ~$10–$39/seat/mo |
| Otter | Cloud (bot) | Otter's servers | Yes — trains on de-identified recordings | Free tier; paid plans |
The fully local three
Speechmark — native, keeps the audio, one-time price
Speechmark records mic and system audio (no bot), then transcribes, separates speakers, and summarizes entirely on-device. Two things set it apart even within the local tier: it keeps the original audio alongside the transcript so you can replay exact wording, and it exposes your meeting history to Claude and other assistants through a local MCP connector. Summaries default to on-device models; you can optionally plug in your own API key, in which case only transcript text — never audio — is sent. $79 one-time per Mac with a free tier. Pick it if you want local processing with a native-app experience and no subscription.
Meetily — free, open source, cross-platform
Meetily (MIT license, ~26,000 GitHub stars, v0.4.0 as of July 2026) transcribes and summarizes locally on macOS, Windows, and Linux, with Whisper/Parakeet models and summaries via Ollama or your own API keys. It's the strongest answer for "free and auditable" — you can read the source. The trade-off is assembly: you choose and configure the models, and polish trails a paid native app. A PRO tier is planned. Pick it if you want open source, run Linux/Windows too, or need to show an auditor the code.
MacWhisper — transcription workhorse, €64 once
MacWhisper is transcription-first: drag in any recording, or capture meetings and app audio directly, all processed locally. Pro (€64 one-time) adds speaker recognition, batch work, and AI summaries. It's less a meeting-notes system than a superb local transcription tool — no meeting-centric organization or assistant integration. Pick it if your workflow is "transcribe this file/call, give me text" rather than a searchable meeting memory.
The bot-free cloud tier: Granola and Fathom
Granola is the best-designed notetaker in the category — menu-bar native, no bot, blends your typed notes with the transcript. But processing happens on Granola's servers, and its pricing page lists a team-wide training opt-out only on the $35 Enterprise tier (as of July 2026). Fathom has a genuinely generous free tier; its privacy policy says it trains in-house models on de-identified meeting content by default, with an opt-out in account settings. Both are fine tools whose architecture — not behavior — is the privacy question. Our full comparisons: Speechmark vs Granola and the Granola alternatives roundup.
The cloud-bot tier: Fireflies and Otter
Credit where due: Fireflies' privacy policy states unconditionally that customer meeting content is not used to train internal or external AI models — the cleanest training stance of any cloud tool here (as of July 2026). It remains a bot-first, cloud-resident team platform, which is the trade. Otter combines a visible bot, cloud storage, a policy of training on de-identified recordings, and the pending class action above. Details in Speechmark vs Fireflies and Speechmark vs Otter.
Where a cloud notetaker is the better choice
Honesty section. Pick a cloud tool when: your team needs one shared, searchable archive with permissions (Fireflies is built for exactly this); you need mobile and web access to every meeting from any device; you want zero-cost capture at volume (Fathom's free tier is real); or you live in CRM integrations — local tools can't push notes into Salesforce workflows the way cloud platforms do. On-device tools are single-Mac by design; if that's disqualifying, the cloud tier graded above is the right shelf.
The new local wave (worth watching, not yet graded)
The on-device thesis is being validated by a wave of new Mac apps — Mumble, MeetMemo, QuickQuill, Nomedo, Hoeren, Slipbox, and others all pitch local processing as of July 2026. Most are early-stage and iterating fast, so we've kept them out of the graded table rather than pin claims to moving targets. One cautionary tale about this segment's churn: Hyprnote, an open-source local notepad we tracked for this roundup, rebranded to Char in mid-2026 and now positions as a cloud-connected "AI chief of staff." Local-first positioning can change under you — one more reason the durable test is architecture you can verify today, not marketing category.
FAQ
What is the most private meeting notetaker for Mac?
The strictest options process everything on your Mac and never upload audio: Speechmark ($79 one-time, keeps the original audio, native app), Meetily (free, open-source, MIT-licensed), and MacWhisper (€64 one-time Pro, transcription-first). Bot-free cloud tools like Granola and Fathom are more discreet than a bot but still process your meetings on their servers.
Is bot-free the same as private?
No. Bot-free means nothing visibly joins your call — but as of July 2026, bot-free tools like Granola and Fathom still upload your audio for cloud processing. If your requirement is that meeting audio never leaves your Mac, the tool must be on-device, not just botless. That distinction is the single most useful filter when comparing notetakers.
Do AI notetakers train on your meeting recordings?
Some do. As of July 2026, Otter's privacy policy says it trains its proprietary AI on de-identified audio recordings, and Fathom trains its in-house models by default with an account-settings opt-out. Fireflies states unconditionally that it does not train on customer data. Fully local tools have nothing to train on — your audio never reaches them.
Are there free private meeting notetakers for Mac?
Yes. Meetily's Community Edition is free and open source (MIT license), and MacWhisper's free tier transcribes locally. Speechmark has a free tier with a $79 one-time upgrade. The trade-off with free local tools is usually setup effort and polish — Meetily requires configuring transcription and summary models yourself.
Try the on-device option first: download Speechmark free — record one real meeting and see the transcript, speakers, and summary appear without anything leaving your Mac.