Chapter 05 · Our Favourite Skills · Reading 06
OpenWhispr, free local transcription
The open-source dictation and transcription app whose Whisper engine we use to get word-level transcripts without sending audio anywhere. By the OpenWhispr team, MIT licensed.
What it is
Every cut in our editing pipeline starts from a word-level transcript, and we did not want to upload raw footage to a transcription service to get one. OpenWhispr is a free, open-source alternative to tools like WisprFlow and Granola: voice-to-text dictation, meeting transcription and notes, running locally on macOS, Windows and Linux. Under the hood it drives whisper.cpp, and that engine is what our HyperFrames transcribe step calls. Audio stays on the machine; the transcript comes back with a timestamp on every word.
Author: the OpenWhispr team. MIT licensed.
Repository: github.com/OpenWhispr/openwhispr
The skills inside it
OpenWhispr is an app first, but it ships two agent skills so an assistant can drive it:
- openwhispr-cli. Operate on notes, folders, transcriptions and audio from a terminal through the
openwhisprcommand (npm package@openwhispr/cli). - openwhispr-api. Build integrations against the OpenWhispr REST API: manage notes and transcriptions programmatically, or read team-space content with a workspace key.
Install it
Download the desktop app from the releases page. For the command line:
npm install -g @openwhispr/cliIf all you need is the transcription engine, whisper.cpp on its own is enough. We keep the small.en model cached locally; it runs at roughly real time on a CPU, which is fine for a sixty-second reel.