There is a claim that has quietly become standard in AI meeting tools: no bot joins your call. It is a real advantage and it is worth having. It is also routinely mistaken for a different claim — that your conversation stays on your computer — and those two things are not the same.

A tool can capture audio locally, put nothing in the participant list, and still upload every second of that audio to a transcription service somewhere else. From the meeting's point of view, nothing joined. From your data's point of view, it left the building.

If you have ever had to answer a procurement questionnaire, or signed an NDA that names sub-processors, you already know why the distinction matters.

Three things people mean by "private"

When a meeting tool says it is private, it is usually making one of three quite different promises.

No bot in the call. Nothing appears as a participant. This is about perception and consent — how the meeting feels, and whether a client sees a third party in the room. It says nothing about where the audio goes.

Local capture. The recording happens on your machine rather than on a server that dialled in. Better, but capture and processing are separate steps, and most tools capture locally then process remotely.

On-device processing. The audio is turned into text by a model running on your own hardware. Nothing is uploaded to be transcribed. This is the only one of the three that means your words never leave your computer.

Most bot-free tools deliver the first two. Considerably fewer deliver the third.

Where the popular tools actually sit

Granola is the best-known bot-free option, and it genuinely does capture on your Mac without joining the call. Its audio is then passed to a cloud transcription provider, transcripts are held on US-hosted infrastructure, and on non-Enterprise plans your data is used to improve its models unless you opt out. None of that is hidden, and for a lot of teams it is entirely acceptable — but it is a different promise from on-device.

Bot-based tools such as Otter and Fireflies are clearer about this by design: the bot joins, the audio goes to their cloud, and the archive lives there. That is the trade, and it buys you good search and mature integrations.

At the other end, fully local open-source options exist and will transcribe entirely offline. They are the strictest choice available, and the cost is usually polish, integrations and someone on your team maintaining them.

If you want the wider field, our comparison of the best AI meeting notes tools covers more of them, and Granola alternatives goes deeper on the bot-free set specifically.

How Lunar handles it

Lunar is a desktop app that captures your microphone and your computer's system audio directly, so nothing joins the call on Zoom, Google Meet, Teams or anywhere else.

Transcription can run fully on-device using a local model. When you choose that, the audio is never uploaded to be transcribed — the text is produced on your hardware. Sessions are stored on your machine and stay there until you sign in; sign in and they sync so the rest of your team can read them.

Some features do use the cloud. AI summaries and the live in-meeting chat call out to cloud AI today. We would rather write that sentence plainly on a page like this than let "private by design" do the work of hiding it. The security page shows the exact data flow per feature, and Lunar is built by an EU company — which matters mostly because it determines who can be compelled to hand over what.

When on-device is worth the trade

On-device transcription is not free. Local models are typically a little slower, and accuracy on heavy accents or bad audio can trail the best cloud models. So it is worth being honest about when that trade is right.

It is worth it when: you are covered by an NDA that restricts sub-processors; you work in healthcare, legal or finance; your client's security review asks where audio is processed; you are subject to an EU data policy and cannot justify US-hosted transcripts; or you are discussing an acquisition, a layoff, or anything else where a leak is not a theoretical harm.

It is probably not worth it when: your meetings are internal and routine, transcription speed matters more than provenance, and nobody will ever ask you the question. Plenty of teams are in this position and should use whatever is most accurate.

The useful thing is having the switch, rather than discovering after twelve months of use that the choice was made for you.

The question to ask any vendor

If you take one thing from this page, make it this. Do not ask whether a tool is private, or whether it is local, or whether a bot joins. Ask:

When I record a meeting, does the audio leave my machine — and if so, which provider receives it, where is it stored, and is it used to train anything?

That single question separates the three promises above, and most marketing pages will not answer it directly. The ones that do are usually the ones worth trusting.

Frequently asked questions

Is bot-free the same as private?

No. Bot-free means nothing joins your call as a participant. Your audio can still be uploaded to a cloud service for transcription. They are separate properties and should be evaluated separately.

Does on-device transcription work offline?

Local transcription does not need a connection to turn speech into text. Features that call cloud AI — summaries and live chat — do need one.

Is local transcription less accurate?

Often slightly, particularly with poor audio or strong accents, though the gap has narrowed considerably. If accuracy matters more than data residency for your meetings, use the cloud option; the point is that it is your choice per your situation, not a decision made for you.

Will my client's IT team accept this?

They will ask the question in the box above, and on-device transcription lets you answer it with "it does not leave the machine". That is a materially easier conversation than explaining a sub-processor chain, which is why agencies and consultants tend to care about this more than internal teams do.

What happens to the recording afterwards?

Sessions live on your machine until you sign in. Once signed in they sync so the team can read them, and the meeting's action items can be pushed to Jira — covered in turning meeting action items into Jira issues.

In short

"No bot" is a claim about the meeting. "On-device" is a claim about your data. Vendors have a strong incentive to let you hear the second when they have only built the first — and the gap between them is exactly the thing your client's security questionnaire is designed to find.