"Local AI" sounds like a marketing word. It isn’t — it’s a real architectural choice with real consequences. Here’s what it means inside Kargoo, and why every word of every meeting genuinely never leaves your Mac.
When you install Kargoo we download Whisper — OpenAI’s open-source speech model — to your disk. After that one download, it never phones home. The model file is yours; transcription runs against it locally.
Kargoo runs Whisper on your Neural Engine. A 60-minute meeting transcribes in 2–4 minutes on M2, faster on M3/M4 — without slowing down anything else. No server queue.
Audio is .m4a. Transcript is .txt. Summary is .md. They sit in a folder you choose. Open them in any editor.
| Local (Kargoo) | Cloud (everyone else) | |
|---|---|---|
| Where your voice is processed | Your Mac’s Neural Engine | A datacenter |
| Number of services that "see" your audio | 0 | 2–5 (recorder, transcriber, summarizer, storage, CDN) |
| Data-breach exposure | Your single machine | Every service in the chain |
| Subpoena exposure | Yours alone | The vendor’s compliance team |
| Used for AI training | Cannot be — we don’t see it | Sometimes (read the ToS) |
| Works on a plane | Yes | No |
That’s the entire list. No engagement analytics, no audio "improvement" pipeline.
Free to download. Free 14-day trial. No account required.
Download for Mac · Free