No bot · no waiting room · on-device
Nobody has to let it in.
Meetings that write themselves.
Tired of another AI agent barging into your call — or waiting for the organizer to admit your notetaker? Vlocalhost never joins the meeting. It runs on your machine, wakes only when someone speaks, and hands you clean notes. Nothing to approve. Nothing uploaded.
No bot. No waiting room.
Stop asking permission to take notes.
Tired of another AI agent barging into your call? Tired of pinging the organizer to admit your notetaker while everyone waits? Vlocalhost never joins the meeting. It runs on your machine and listens to you — invisible in the participant list, nothing to approve. Just talk.
Already in. Because it never left your side.
No join request. No “recording has started.” No one in the room even knows it’s there — but you walk out with the notes.
Switch to Vlocalhost →The signal path
Five stages. Zero cloud.
Audio enters at the microphone and never leaves the device. Each stage runs locally — capture, gating, transcription, and reasoning — so the only thing that reaches disk is a finished set of notes.
Capabilities
Engineered around one idea: listen only when it matters.
Silence is ignored
The engine only transcribes genuine speech. Coughs, keyboard clicks, pauses, and dead air never reach the model — so transcripts stay clean and fast.
Nothing leaves your device
Transcription and summarization both run on your own hardware. No API keys, no accounts, no telemetry, no cloud round-trip. Ever.
It knows your meetings
Connect Google or Outlook and Vlocalhost names every file from the real event, and can start recording the moment a meeting begins.
Runs on everything
Windows, macOS, and Linux — the same engine, the same privacy guarantees, one command to launch. Tray app or terminal, your call.
No model limitation
Bring your own voice model.
Vlocalhost never locks you to one engine. Attach the speech-to-text model that fits your accent, your language, and your hardware — swap it in a single line, or replace the engine entirely. Small and fast, or large and pin-sharp. Your model, your rules.
Any Whisper model
From tiny.en to large-v3, or any fine-tuned model on Hugging Face — one line in config.py.
A local model folder
Point Vlocalhost at a model on disk. Any size, any language, fully offline — nothing to download.
Your own engine
Plug in a custom transcriber with a .transcribe() method and bypass Whisper completely.
Calendar + delivery
Connected when you want it. Offline when you don't.
Auto-start on meetings
Vlocalhost watches your calendar and begins recording as the meeting starts.
Name notes from the event
Files are titled from the real meeting, not a guess.
Email the summary
Send finished notes to attendees the moment the meeting ends.
Post back to the event
Write the notes straight into the calendar event's description.
Credentials cached in your OS keystore folder — never in the cloud, never in the repo.
Instrument spec sheet
The numbers under the hood.
| Capture | 16 000 Hz · mono · 16-bit PCM |
|---|---|
| Voice gate | webrtcvad · 30 ms frames · aggressiveness 2 |
| Silence timeout | 800 ms trailing silence ends an utterance |
| Transcription | faster-whisper · base.en · int8 / CPU · bring your own voice model |
| Summarization | Ollama · llama3.2 · fully local · bring your own model |
| Output | Markdown notes + timestamped .txt transcript |
| Platforms | Windows · macOS · Linux |
| Network egress | none — 0 bytes off-device |
Install
From silence to notes in one command.
Clone it, point it at a local model, and talk. The first run pulls the Whisper weights once, then everything is offline for good.
Download Vlocalhost →# set up an isolated environment $ python -m venv .venv $ source .venv/bin/activate # .venv\Scripts\activate on Windows $ pip install -r requirements.txt # pull a local model for summaries $ ollama pull llama3.2 # start listening — silence is ignored $ python vlocalhost.py 🎤 Vlocalhost online · on-device · watching for speech