No logo wall here yet. When companies we can name are using it they will go in this row — until then it carries things you can check instead.
Why Vlocalhost
Tired of an AI notetaker barging into the call — or waiting for the organiser to admit it? Vlocalhost never joins. It runs on your machine, wakes when someone speaks, and hands you the notes when you stop.
Nothing to approve and nobody to admit. The other people on the call never see a third participant, because there isn’t one.
It starts when you start. No host to let it in, no queue, no meeting link to paste anywhere.
Audio, transcript and notes are written to your own disk and stay there. There is no server to send them to.
Audio enters at the microphone and never leaves the device. This is the real pipeline, running end to end — capture, gate, transcribe, structure, deliver.
Six layers, one process, no network boundary between any of them. Scroll to pull them apart; hover a layer to isolate it.
Attach the model that fits your accent, your language and your hardware — swap it in a single line, or replace the engine entirely. Pick one to see the trade.
A timestamped transcript on the left, structured markdown on the right, both written to disk under the meeting's real name.
A 30 ms frame either contains speech or it doesn't. Only the yellow frames ever reach Whisper — coughs, keystrokes and dead air are dropped before they cost you a cycle.
Every packet the engine produces terminates inside the machine boundary. No API keys, no accounts, no telemetry. Calendar sync, if you switch it on, is the one outbound call the app can make — and it carries no audio, transcript or note.
Every notetaker on the market says your data is safe. This is the part where you check ours — none of it asks you to trust the sentence above.
Switch on airplane mode and run a whole meeting. Capture, transcription, summary and the file on disk all still work — there is nothing for the network to do.
Point a packet monitor at the process while it records, and watch it stay quiet for the length of the meeting.
The whole audio path is three files in the open core, in that order. It is a short read, and there is no upload in it.
Point Claude or ChatGPT at the repository and ask it what the app sends. We would rather you asked something impartial than believed us.
The same engine and the same privacy guarantees on each — one command to launch. Tray app or terminal, your call.
Defaults on first run. Every one of them is a line in config.py.
Every notetaker draws the first box the same way and stops there. This is the rest of the diagram — not who joins your call, but where your voice ends up once it has.
Everyone in the meeting sees it arrive, and the recording starts its journey somewhere else.
The participant list stays clean. The audio takes exactly the same trip as it did before.
The same pipeline, with the network taken out of it. There is no next hop to draw.
Free · Solo plan
Tell us where to reach you and we’ll take you straight to the installer. No card, no account.
