Why we run Whisper on your machine

Most language-learning tools that transcribe video upload your file to a server, run the transcription there, and send the subtitles back. It’s the easy path: their engineers don’t have to worry about your CPU, your operating system, or the size of your library. The cost is yours: you wait for the upload, you depend on their pricing, and your private film collection sits on a stranger’s hard drive for as long as their retention policy says.

Octopus does the opposite. When you drop a film in, the transcription runs locally on your computer using Whisper, the same model that powers professional transcription tools. It never leaves your machine.

The numbers

We tested a 40-minute BBC documentary on a recent laptop. Whisper finished in 2 minutes 9 seconds — about 18× faster than playback. A 90-minute film takes roughly 5 minutes. A short YouTube clip is done before you’ve made coffee.

Going local means:

• No upload time. A 4 GB film transferring to a server at 50 Mbps takes ~10 minutes before transcription even starts. Locally, transcription begins instantly.
• No per-minute fees. Pay-as-you-go transcription APIs run $0.006–$0.024 per minute. A 90-minute film costs $0.50–$2.00 every time you transcribe it. Octopus is a one-time license, no usage charges.
• No retention policy to read. Your films are read on your machine, the transcript is saved on your machine, and we never see either.

What runs locally, what doesn’t

To be honest about it: there’s exactly one thing that goes online, and only when you ask. When you click Translate this sentence, that single sentence is sent to the translation API endpoint you configured. Octopus doesn’t proxy it through our servers. The request goes from your machine straight to your translation provider, governed by their privacy policy, not ours.

Everything else — your film files, your transcripts, your vocabulary, your reading progress — stays on disk.

Why this matters for language learners

If you’re learning a language seriously, you’re going through dozens of films and books over months. The cost of doing that with a cloud-based tool adds up — both in dollars and in the slow erosion of having a third party know exactly what you watch and read. Local-first isn’t a marketing line for us; it’s the only way to make the math work for the kind of long-term, library-scale study that actually moves the needle.