Scribe

Record. Transcribe. Summarise.
← Back to Scribe

How It Works

Your recordings never leave your device.

Client-Side Processing

Scribe captures and transcribes audio entirely inside your web browser. When you record a session, the audio is held in browser memory and processed by a machine-learning model that runs locally on this device. Nothing is uploaded, streamed, or stored outside your browser.

What Stays Local

  • The microphone stream — every byte of audio.
  • The Whisper speech-recognition model (cached in your browser after first download).
  • The pyannote speaker-diarization model, if you've enabled it.
  • The generated transcript (until you choose to export or send it).
  • Your saved session history (stored in your browser's IndexedDB).

What Leaves Your Device (Only When You Ask)

  • Summarisation. If you click Summarise, Scribe sends the transcript to Anthropic's Claude API via our authenticated proxy. You must sign in with your @gscls.com.au Microsoft account first. Nothing is sent until you click the button.
  • Actionstep export. If you click Send to Actionstep, the transcript and summary go through the organisation's Logic App endpoint into the matter you select. This only runs on explicit user action.

Recording and transcription never leave the device — those steps work offline, with no sign-in.

How It Works

1

You open Scribe in your browser. The web app (HTML, CSS, JavaScript, ML models) is downloaded once and cached for future visits.

2

You grant microphone access and record. The audio stream is captured into browser memory — it is never uploaded.

3

Whisper runs locally in your browser, transcribing the audio chunk-by-chunk using WebGPU or WebAssembly.

4

The transcript appears in the panel. You can export it, copy it, or discard it — it stays on your device until you explicitly share it.

Installable App

Scribe is a Progressive Web App. You can install it on your phone or desktop from your browser's menu. Once installed, it works even when the internal server is temporarily unavailable — the application code is cached locally.

Data We Don't Collect

  • No analytics, tracking, or telemetry.
  • No third-party cookies.
  • No audio, transcript, or summary data is retained on our servers.
  • Your Microsoft sign-in only grants access to the summarisation endpoint; we do not store your profile.

Open-Source Libraries

Scribe is built entirely on open-source software:

  • Transformers.js — ML inference in the browser (Apache 2.0)
  • OpenAI Whisper — speech recognition (MIT)
  • pyannote-audio — speaker diarization (MIT)
  • ONNX Runtime Web — model execution (MIT)
  • PDF.js — agenda PDF parsing (Apache 2.0)
  • Fira Sans — typography (SIL Open Font License)