Home · Private meeting notes for Mac

Private Meeting Notes for Mac: How to Record Without Bots

"Private" is the most overloaded word in the meeting-notes category. Every cloud notetaker claims it; almost none of them mean what individual lawyers, therapists, journalists, or compliance-bound professionals mean by it. This guide walks through the actual threat models, what each kind of app does and doesn't protect, and how to take meeting notes on a Mac without anything leaving your machine.

What "private meeting notes" actually has to defend against

Before evaluating any tool, get specific about what privacy means in your context. Four distinct threat models show up in real conversations:

  • Other meeting participants seeing that you're recording. Cloud notetakers usually solve this poorly: they send a visible bot ("Otter Assistant", Fireflies' "Fred", MeetGeek's bot) into the call, which creates social friction and sometimes legal questions in two-party-consent jurisdictions.
  • Third-party SaaS vendors having your audio. Otter, Fireflies, Granola, MeetGeek, Fathom, Sonix all upload audio to AWS or GCP. Their employees, infra, and incident-response teams are in the trust boundary. Their breach disclosures are your problem.
  • Subpoena and legal-discovery exposure. If a third party holds your audio, that third party can be subpoenaed. For lawyers and journalists this is a non-trivial risk.
  • "Trained on" and AI-feature scope creep. Even when current ToS says "we don't train on your data," the next version might. Anything in a vendor's storage is one policy update from being in their model.

An honest privacy story has to address all four. Most cloud apps address #1 and #4 with policy language and ignore the rest.

The three ways to take notes from a meeting on Mac

1. Cloud notetaker with a bot

How it works: you authorize the app to access your calendar, it joins meetings as a bot, audio goes to AWS, transcript and summary come back. Examples: Otter, Fireflies, MeetGeek, Fathom. Convenient, broad platform support, team-friendly. Worst privacy story.

2. Cloud notetaker without a bot

How it works: a native client records audio locally, then uploads it to the vendor's cloud for transcription and summarization. Example: Granola. Better than option 1 (no visible bot) but the audio still leaves your Mac. The privacy delta vs option 1 is smaller than the marketing suggests.

3. On-device transcription with no upload

How it works: a native macOS app uses ScreenCaptureKit to capture system audio + mic, then runs Whisper (via WhisperKit on Apple Silicon, whisper.cpp on Intel) entirely on your Mac. Audio is stored as a local file. Transcript is stored in a local database. Nothing leaves the device unless you explicitly send a transcript out for AI summarization. This is the path TalkFold takes. See the full offline transcription guide for the technical detail.

How to record Zoom, Google Meet, and Microsoft Teams without a bot

The technical primitive that makes bot-free recording possible on modern macOS is ScreenCaptureKit, Apple's framework for capturing screen and audio output natively. A native app that uses ScreenCaptureKit can capture the audio stream of any other app — Zoom, Google Meet (in Chrome/Safari), Teams, Slack huddles — together with your microphone, into a single audio file.

Concretely, the flow looks like this:

  • Click record in your notetaker app (or hit a global hotkey).
  • The app requests Screen Recording permission once. macOS prompts you the first time; after that it's transparent.
  • Join your Zoom / Meet / Teams call as normal. Speak normally. There is no extra participant in the call.
  • Stop recording when you're done. Transcription runs locally afterward (or in parallel, depending on the app).
  • The original audio sits as a WAV in your local notes database. Nothing is uploaded.

From the perspective of the other call participants, the meeting is identical to a meeting where you're just typing notes. There is no notification, no bot avatar, no "Recording in progress" banner created by your tool.

What data still leaves your Mac (be honest)

Even on the strictest on-device app, "nothing leaves your Mac" is rarely literally true. A useful disclosure looks like this — TalkFold's data flow as a worked example:

  • Audio: never leaves the device. WAV files in the app's local storage, deleted when you delete the recording.
  • Transcripts: never leave the device automatically. Stored in a local SQLite database. They go out only if you explicitly export, copy/paste, or trigger an AI summary.
  • AI summaries (Pro): the transcript text — not audio — is sent to a hosted LLM via TalkFold's API only when you click "Summarize." Audio is never part of that request.
  • Email Groups (Pro): when you choose to email a summary, the recipient list is passed to Resend (a transactional email provider) for that delivery. The list is not stored after delivery.
  • Subscription metadata: for Pro users, your Ko-fi subscription email, your license key, and a monthly request count exist on TalkFold's server. No transcripts, no summaries, no audio, no dictionary, no user-generated content.

This kind of itemized accounting is what "private" should mean. If a vendor can't give you the same itemized list, the privacy claim isn't verifiable.

Practical setup for a privacy-first workflow on Mac

For lawyers and paralegals

Use an on-device app for client calls and depositions. Keep recordings locally on a FileVault-encrypted Mac. If the matter is privileged, do not run AI summaries through any external service — TalkFold's free tier covers this case (transcription only, no summaries).

For therapists and clinicians

Same model. The free tier of an on-device app gives you a local transcript without invoking HIPAA at all, because no PHI ever crosses the device boundary. Treat the local SQLite database as PHI storage and apply the same FileVault + access-control hygiene as your case notes.

For journalists

Record source interviews on-device. If you need a summary, run it only after redacting source-identifying details. Better yet, do summaries by hand for the most sensitive interviews — the marginal value of an AI summary is rarely worth the source-protection risk.

For compliance-sensitive roles

An on-device app is much easier to get through internal review than a cloud SaaS, because the security review only has to evaluate one machine, not a vendor's entire SOC 2 program.

Comparing tools that claim "private"

Each of these has a head-to-head page covering exactly what their privacy story does and doesn't include:

Get started with TalkFold

If everything in this guide describes the trade-offs you actually care about — no bot in calls, no audio in the cloud, AI summaries strictly opt-in and transcript-only — TalkFold is the path.

  • Free tier: unlimited recordings, transcription, dictionary (5 words), shortcuts (5). No account.
  • Pro: $4.99/month or $49.99/year — adds AI summaries with 5 templates, Email Groups, unlimited dictionary and shortcuts.
Download TalkFold Free See Pricing

Requires macOS 14.0 Sonoma or later. Free forever.