A local meeting recorder for Mac: what “local” should mean, and how to check

Almost every meeting tool now says your data is private. Some mean the audio never leaves your computer; some mean it is encrypted on their servers; some mean they do not train on it. Those are very different promises. If you are looking for a meeting recorder that actually runs on your Mac, five questions sort them quickly.

1. Where does the audio become text?

This is the question that matters most, because audio is the most sensitive thing a meeting produces. A local recorder transcribes on your Mac. A cloud one uploads the audio and sends back text, however private its marketing sounds.

Daisy transcribes on the Mac with Whisper on the Apple Neural Engine. The speech model is downloaded once, on first run, from huggingface.co, and cached on your disk. After that, recording and transcription work with the network off.

2. Where is the summary made — and do you get to choose?

Summaries need a language model, and the good ones are often in the cloud. What matters is whether you choose, and what exactly is sent.

In Daisy you pick the summarizer: Apple Intelligence on the Mac (macOS 26), a local model through Ollama or LM Studio, or a cloud provider — OpenAI, Anthropic, Kimi — with your own key or ChatGPT account. With a cloud provider, the transcript text goes from your Mac straight to that provider under its terms, never audio and never through Daisy's servers. An optional on-device filter can replace names, companies and contacts with placeholders before it leaves.

3. Where does the archive live, and in what format?

A recorder that keeps your meetings in its own database, or in its own cloud workspace, owns your archive in practice even if it is technically on your disk.

Daisy writes one folder per meeting into a folder you choose — an Obsidian vault, a Dropbox folder, anywhere: transcript.md (Markdown you can open in any editor), summary.json, and the audio as .caf files. There is no database only Daisy can read. You can delete the audio after transcription and keep the text.

4. Does it need an account, a bot, or a cloud workspace?

Each of those is a sign that your meetings pass through someone else's system. A bot that joins the call is audio on a vendor's server by definition.

Daisy has no account, no bot, no Daisy cloud archive, no telemetry and no analytics SDK. It records the call from your Mac's own audio, so nobody else sees a participant join.

5. Can you check any of this yourself?

A privacy claim you cannot test is a claim. Two things make it testable: a network test anyone can run, and source code anyone can read.

For Daisy, the test takes about ten minutes: block it in a firewall such as LuLu or Little Snitch, set the summarizer to Apple Intelligence or a local model, record a real meeting, and see that the transcript and summary still appear. The verify-local page lists every host the app can contact and what triggers each one. The source is public under Apache 2.0.

What “local” does not cover

• Update checks go to mydaisy.io, hourly by default. They carry no meeting data.
• Speech models download from huggingface.co the first time you choose them.
• A cloud summarizer, if you choose one, receives the transcript text.
• Destinations you connect — Notion, Slack and others — receive what you route to them.
• iPhone sync, off by default, goes through your own iCloud, not to us.
• Recording laws are yours to follow: some places require every participant's consent.