How Alder is built.

We assumed one universal model would be enough. Measurement said otherwise — so Alder picks its on-device engine per language, and users never think about it.

Running a speech model once is easy. Delivering multi-gigabyte models to thousands of different Macs — and keeping them healthy — is the real product.

Bluetooth microphones that never deliver audio, quit-mid-sentence sessions, text fields that refuse insertion: how Alder is engineered so a spoken thought always survives.

A clear guide to Alder’s Mac-native, end-to-end on-device dictation pipeline, privacy boundary, requirements, and pricing.

Six checks that reveal whether audio, transcripts, and cleanup actually stay on your Mac.

Built for 中英混说: native punctuation, one script, optional cleanup, and local Chinese–English translation.

Measured on 日英混在 speech: correct 句読点, English kept as English, and local Japanese–English translation.

Every accent and ¿¡ in place, Spanglish handled natively, and local Spanish–English translation.

한영 혼용 kept as spoken, optional cleanup, and local Korean–English translation on the device.

Benchmarked directly: Umlaute and ß correct, Denglisch kept as spoken, local German–English translation.