Why I don't trust my meeting notes to the cloud anymore

What's actually in your meeting notes

Think about your last ten meetings. Not the all-hands or the team standup. The other ones. The ones where you discussed who's getting promoted and who isn't. The compensation review where you talked about specific salary numbers. The client strategy session where you mapped out competitive positioning. The board meeting where you shared runway numbers. The 1-on-1 where someone told you they're thinking about leaving.

Now think about where those recordings went.

If you're using Otter.ai, Fireflies, Granola, Fathom, or any other cloud-based meeting tool, the answer is: to a third-party server. Your audio was uploaded, processed, stored, and indexed on infrastructure you don't control, can't audit, and can't delete from with certainty.

This isn't hypothetical risk. This is the default behavior of every major meeting transcription service.

The Granola MCP moment

In early 2026, Granola made a product decision that crystallized the problem. They removed their local Obsidian sync — the feature that let users keep meeting notes on their own machines — and replaced it with a cloud MCP server. The pitch: your AI agent can now query your meetings through Granola's API.

But read that again. Your meeting data now flows through Granola's servers to reach your own tools. The data that used to live on your filesystem now requires a cloud intermediary to access. That's not a feature improvement. That's a custody transfer.

Granola isn't doing anything unusual here. This is the standard playbook for AI tools: collect user data, process it centrally, then monetize the processing layer. The meeting audio is the raw material. The cloud API is the toll booth.

The question you should be asking

Every time you use a meeting recording tool, ask one question: where does my audio go after I press record?

If the answer involves the word "server," "cloud," "API," or "upload" — your audio leaves your machine. Once it leaves, you lose control over it. You're trusting that:

The company encrypts it properly in transit and at rest. The company doesn't use it for model training. The company's employees can't access it. The company won't be breached. The company won't change their privacy policy. The company won't be acquired by someone with different values. The company will actually delete it when you ask.

That's a lot of trust for a tool that records your most sensitive conversations.

The trend that should worry you

It's not just meeting tools. The entire AI ecosystem is converging on a model where your data flows to someone else's infrastructure for processing. Char depends on OpenRouter. Granola depends on their cloud. Every "AI-powered" tool that processes your data remotely is creating a copy of your information on systems you don't own.

The argument is always convenience. "We need the cloud for processing power." "The models are too large to run locally." "The experience is better when we can process server-side."

For meeting transcription, none of these arguments hold up anymore. Apple's Neural Engine in M-series chips runs speech-to-text models locally, in real time, with accuracy that matches or exceeds cloud APIs. The processing power argument expired when Apple shipped the M1. There is no technical reason your meeting audio needs to leave your Mac.

Local-first is not paranoia

I want to be clear: I'm not suggesting that Granola or Otter are malicious. I'm not suggesting they're selling your data or that their security is bad. I have no evidence of that, and I assume they take security seriously.

But security isn't the only risk. There's also the structural risk of dependency. When your meeting history lives on someone else's servers, you depend on them to maintain access. If they change their pricing, you pay or lose access. If they change their API, your workflow breaks. If they shut down, your data may or may not be exportable. If they get acquired, new owners set new terms.

Local-first eliminates all of these risks. Not by being paranoid, but by removing the dependency entirely. Your files are on your machine. They work when you're offline. They work if the company goes bankrupt. They work with any tool that can read a text file.

That's not paranoia. That's good data hygiene.

The case for keeping meetings on your Mac

Transcripted is a free, open source macOS app that records and transcribes meetings entirely on your Mac. Audio is processed by Apple's Neural Engine — it never touches a server. The output is plain markdown files in a folder you control.

No account. No upload. No cloud processing. No subscription. No MCP server sitting between you and your data.

When I record a compensation review with Transcripted, the audio is processed on my Mac and the transcript is saved to my filesystem. The audio data exists in one place: my machine. I can encrypt my disk. I can back it up however I want. I can delete it with certainty. No one else has a copy.

This isn't a limitation. It's the point.

What changes when your meetings are truly private

Something interesting happens when you know your meetings are genuinely private. You stop self-censoring. You stop wondering if you should pause the recording before discussing something sensitive. You stop having the "is it okay if I record this?" conversation because the recording never leaves the room (metaphorically).

You have more honest meetings. More productive conversations. Better notes from those conversations. And a searchable archive of your professional life that belongs to you — not to a startup that might pivot next quarter.

Your meetings are yours. Keep them that way.