I stopped paying for a cloud dictation app and built my own meeting notes. A transcription model and a language model on our own machines turn speech into clean text, with a shared dictionary for the words it gets wrong, and a meeting into a summary with decisions and action items. Nothing spoken leaves the office. What it does, and why that matters.
Speech is the data I type least carefully. I'll say things out loud in a meeting that I would never paste into a chat window: a client's budget, why a project is stuck, how an employee is doing. And most dictation and meeting-notes tools send that audio to someone else's servers to be transcribed, and then to a cloud AI to be summarised.
So I replaced the paid dictation app I used to have, and built meeting notes on top of the same idea. A transcription model and a language model, both running on our own machines. Nothing spoken leaves the office.
What it does, today
Dictation is simple. I speak, and I get cleaned-up text back: the filler words gone, the punctuation in place, the sentence readable. It's the same thing the paid app did for me, minus the cloud.
The dictation has its own dictionary too. Transcription trips over names, jargon and words from another language, like a Dutch word in the middle of an English sentence. I dictated "offerte" and got "offer to". I corrected it once, and it has come out right ever since. The words in the dictionary are given to the transcription as a hint, and its replacements are applied to the text afterwards. It's shared, so the desktop, the laptop and the phone all use the same one. "Factuur", "btw" and "dossier" are in there now.
Meetings work like this. I record on my phone, upload the recording when it's done, and a bit later I have a summary with the decisions that were made and the action items, each with an owner and a deadline. The full transcript sits next to it. Both are stored on our own storage, in a folder per meeting. Once the processing is finished, the audio is deleted. What's left is the text I actually need.
Why it matters that it's local
Think about which conversations you'd want a note of. A client meeting where prices and contracts come up. A review or a difficult talk with someone on your team. A call with your accountant or a lawyer. That is personal data and business-sensitive information, and under GDPR you stay responsible for it, also when a tool processes it for you.
With a cloud tool, a recording of that conversation sits on a provider's servers, gets transcribed there, and the text gets read by an AI that also belongs to someone else. You can sort that out with a processing agreement and clear retention terms, and sometimes you should. With the local version the question mostly disappears: no third party hears the audio, no third party reads the transcript, and the recording is gone as soon as the notes exist.
What it means for your company
You don't need our setup to take something from this. Look at which of your recurring conversations you'd never want on someone else's server: HR talks, client calls with numbers in them, management meetings. Those are the ones where a local transcription and summary is worth looking at. The meetings that contain nothing sensitive can keep using whatever is convenient.
I haven't built this for a client yet. It's my own setup, in my own office.
What about you?
Which conversation in your company would you never want a recording of on someone else's server? Tell me which one it is, and I'll come and look. The first look is free, no strings.

