On your device or in the cloud: two ways to transcribe a conversation
When you set your phone on the table in a board meeting, a client negotiation or a doctor's appointment, the most sensitive thing in the room is not the summary. It is the raw audio.
Most note takers give you one answer to where that audio goes: their servers, every time. A few privacy-minded tools give the opposite answer and treat the cloud as something to apologise for. We think both miss the point. A flight to Lisbon and a ninety-minute workshop at your desk are different situations, and they deserve different tools.
So Parley Notes transcribes in two ways, and you choose with one switch.
On your device: the audio stays with you
Download the transcription model once, on your phone or in the Windows or Mac app, and your own device turns speech into text. The audio does not reach our servers. Only the text travels, over an encrypted connection, to be written up as your note.
This is the path for a plane, a basement plant room or a depot with no signal. The transcript is made on the spot and the note follows once you are back online. It is also the path for a room where the rule is simple: no audio leaves the building.
It has a cost, and we would rather say so. Speech recognition is heavy work for a phone. On a long day of site visits you will see it in the battery, and the first download takes space.
In the cloud there is nothing to set up and nothing is kept
The web app and the browser extensions work without installing a model, and the phone and desktop apps can do the same. The audio travels over an encrypted connection, our AI provider in the EU turns it into text, and we delete the audio as soon as the text is ready. If that ever fails, it is deleted automatically after one day.
It is also the more accurate path. In our own tests on real calls, the cloud made fewer than half the word errors of the model that runs on a phone. And your phone does none of the heavy work, even across back-to-back calls.
The cloud path is a conduit, not an archive. We do not keep your audio on our servers, and neither we nor our AI provider use your audio, your transcripts or your notes to train AI.
Side by side
| On your device | In the cloud | |
|---|---|---|
| Where the audio goes | It stays on your device. | To our AI provider in the EU, then deleted as soon as it is text. |
| Setup | One download of the transcription model. | None. |
| Word accuracy | Close to the cloud on clear speech. | Fewer errors on real calls in our own tests. |
| Who does the work | Your device. | Our servers. |
| Connection | Transcribes without one. | Needs one. |
| Used to train AI | Never. | Never. |
| Best when | You are on a flight, on a remote site or in a room where audio must not leave. | You are at your desk, in the browser or on a long day of calls. |
The same promise on both paths
Whichever way the words become text, the rest is the same. Your notes are stored in the EU. You can delete one note or everything, yourself, at any time. And if you turn on Private transcripts, your transcripts are locked with a key that stays on your devices. We store them, but we cannot open them.
You should not have to choose between a tool that is easy at your desk and one you can trust in the field. Use the switch, and pick the path that fits the room you are in.
From gut feel to calibrated: the habit behind honest forecasts and deadlines
Reps guess deal probabilities in round numbers and project leads plan deadlines from the best case and both break for the same reason.
Where meeting notes go to die
Every team has a folder of AI meeting summaries nobody opens twice and a CRM nobody updates on Friday afternoon and both come from the same problem.
Half your pipeline happens away from your laptop
Meeting bots work fine when everyone sits at a desk on a video call but they cannot help in a parking lot after a client visit or on a mobile call or inside a six-minute voice note.