Privacy.

Carbon records and transcribes on your Mac, with Apple’s on-device speech recognition. It tells speakers apart on your Mac too, with models that ship inside the app, so nothing is downloaded for it.

Each transcript is a plain Markdown file in ~/Carbon, or the folder you choose. The audio is deleted once the transcript is written, unless you turn on Keep audio. I never see any of it.

When you name a speaker, Carbon remembers their voice as a Voice Print, a numeric voice signature. Voice Prints are stored only on your Mac, and Carbon never sends them anywhere. You can delete them, or turn voice recognition off, in Settings.

No account to sign in to, no analytics, and no server of mine for Carbon to report to. Nothing about your meetings, or how you use Carbon, is sent anywhere.

Carbon asks carbon.alexandrejoly.me whether a newer version exists, and downloads it if there is one. That’s Sparkle, and it’s the only call Carbon makes. You can turn automatic checks off in Settings.

The only other traffic is macOS downloading Apple’s own speech models.

By default, Carbon starts transcribing when a call starts, and its menu bar icon turns into a moving mic meter while it does. You’re responsible for telling the others on the call that they’re being recorded.

carbon.alexandrejoly.me sets no cookies and runs no analytics or tracking scripts. It runs on Cloudflare (opens in a new tab), which processes each request, including your IP address and browser details, to serve the page and protect it from abuse. Request logs are kept for a few days to debug errors.

Write to alexandre.joly36@gmail.com, or open an issue on Carbon’s repository (opens in a new tab). If any of this changes, I’ll update this page.