What private has to mean for an AI notetaker
A private AI notetaker comes down to three questions, and each has a short answer.
- Where is the audio transcribed? On your computer, or on a server run by someone else. If the speech is turned into text on your own machine, no recording of the meeting has to be sent anywhere.
- Where is the transcript kept? On your own disk, or in an account held by the vendor.
- What happens to both afterwards? How long the recording is kept, and whether any text is sent away to write a summary.
Here is how Softspoken answers them. A speech model on your Mac or Windows computer does the transcription, so the audio is never uploaded. The transcript and your notes are ordinary files under your own user profile. By default the recording is deleted once your notes are written, and you can choose to keep it longer.
Nothing from your meetings leaves your computer unless you send it yourself. There are two ways to do that: the optional AI layer, which is the step that writes summaries and cleans up dictation, and a connector to an AI tool that you set up. Either way it is text, never audio. The section on what still leaves your computer has the detail.
That is why we call Softspoken a private meeting notetaker for Mac and Windows.
Bot-free is not the same as private
Recording without a bot is now common. Of the six products in the table below, Jamie, Granola, Wispr Flow and Copilot work with no bot in the call. Otter and Fireflies keep a bot as their main path and also offer bot-free recording in their desktop apps. Softspoken has no bot either. It listens from your computer, so nothing joins the call and nothing is added to the participant list.
Bot-free describes how the audio is captured. It says nothing about where the audio goes afterwards. None of the six says it transcribes meetings on your computer. Most of these applications say that the audio goes to a server for that step, either one the vendor runs or one that belongs to a transcription provider it uses. What happens after that varies too. Jamie, for example, says it deletes the audio once it has been transcribed, which is similar to what we do here at Softspoken.
Sending audio to a server is a common way to build a notetaker, and plenty of meetings are not confidential. It is still a different thing from keeping the audio on your machine, and some more recent changes happened in the past month: Wispr Flow brought its bot-free Notetaker to Windows, and Fireflies added dictation to its desktop app on every plan. Both are real gains for the people who use them. In both, the speech still goes to a server to be transcribed. Therefore, if you have confidential and private meetings, or even if you dictate private emails for example, it is important to seek out how this works and where your data flows with existing providers.
So, a notetaker can be bot-free and still send every meeting to the cloud. Whether it's truly private depends on where the audio goes. The table in the next section sets out, row by row, how each product records and where its audio goes.
Where they are level, and where they differ
Each column was updated from each vendor's website and public documentation, as of 7 October 2026. Partly means the feature comes through a phone app, separate hardware, one platform only, or a mode that sits beside the main product.
softspoken. | Otter | Fireflies | Jamie | Granola | Wispr Flow | Copilot | |
|---|---|---|---|---|---|---|---|
| No bot joins the call | Yes | Partly | Partly | Yes | Yes | Yes | Yes |
| Meeting notes included | Yes | Yes | Yes | Yes | Yes | Yes | Yes |
| In-person meetings as a first-class feature | Yes | Partly | Partly | Yes | Partly | Yes | Partly |
| Dictation included | Yes | No | Yes | No | No | Yes | Partly |
| Audio never leaves your device | Yes | No | No | No | No | No | No |
| Works offline | Yes | No | No | No | No | No | No |
| Pay once, own it | Yes | No | No | No | No | No | No |
Where they are level
If your first requirement is no bot in the call, most of the table now meets this requirement. Four of the six work that way outright, and Otter and Fireflies offer it in their desktop apps, beside the bot that is still their main path. All six write meeting notes. This is where everything is equal to some degree.
Where the picture is mixed
In-person meetings are a main feature for Jamie and Wispr Flow, and the other four cover them only partly. Dictation is included by Fireflies and Wispr Flow, partly by Copilot, and not by Otter, Jamie or Granola. Softspoken does both, in the same app as its meeting notes. Fireflies and Wispr Flow now put meeting notes and dictation in one app as well, so that is common ground with them.
Where they differ
The last three rows are where the products part. None of the six says it turns a meeting's speech into text on your computer, and most require online functionality and cannot work offline. Some can record offline and transcribe once the connection returns, as Wispr Flow's Notetaker does. All six are sold by subscription only. Softspoken answers yes to all three of those rows.
What still leaves your computer, and when
On Softspoken, with the AI layer off and no connector set up, nothing from your meetings or your dictation leaves your computer. Recording, transcription, your notes and search all run on it.
With no connection
Because the speech model runs on your machine, recording, transcription and dictation carry on with every connection blocked. A speech model for meetings is downloaded once, the first time it is needed, so that step needs a connection the first time. The AI layer, sign-in and updates wait until you are back online.
With the AI layer on
The AI layer is optional. It writes the summary and action items for a meeting, answers questions from across your meeting library, and cleans up a dictation. When you use it, the text for that one step is sent, and audio never is. For a question across your library, the search runs on your computer and only the passages that match are sent.
You choose where that text goes.
- Softspoken AI. The text is processed in Sydney, Australia, and returned. It is not stored, not logged and not used to train anything.
- Your own keys, on a paid plan. The same step runs on your own Anthropic key or AWS Bedrock account, and the text goes straight to that provider without passing through us.
- Off. You keep the full transcript and write your own summary.
There is one more path, and it is one you set up yourself. Paid plans include a connector that lets an AI tool you choose, such as Claude, read your meeting library. It runs on your computer, and the meeting text you ask about goes to that tool. The answer on connecting AI tools has the detail.
The app has no telemetry. The diagram on How it works shows what is sent under each of the three choices, and the network requirements page has the full list of network destinations, host by host. The security page has the rest.
What it costs
Recording, transcription, notes and search all run on your computer and are free, with no account and no card. That is the Free plan, A$0 forever.
The paid plans are for the AI layer. Pro is A$19 a month, or A$15 a month, billed as A$180 a year. It includes meeting summaries and answers from across your meeting library, with dictation cleanup as well, all processed in Sydney.
Perpetual is A$299, paid once. The licence does not expire and updates are included. The AI layer runs on your own Anthropic key or AWS account.
The pricing page has the full detail, including what the free plan includes for AI.
Softspoken is in closed beta, and invitations go out by email to the waitlist.
More answers
The common questions page covers the rest. Four of its answers bear on confidential meetings: