Hold the key and talk.

Verba is the English-facing build of the same dictation app. One engine, one source tree, and the entire difference between the two builds fits into ninety lines that hold the name, the domain and the support address. The releases, the numbering and the icon are its own.

Naming it took four dead ends. Quill, Plume, Scriba and Vox all fell over on trademark checks. Quill is taken by an editor and a rewriting tool, Plume has a live registration in our own category, Scriba had a fresh filing describing speech to text word for word, and the class Vox sits in is crowded, plus there is a well-known Mac player already using it. Verba came through clean. It is Latin for words.

There is no domain yet, verba.app belongs to a live product. So the app goes out through GitHub instead of a nice address, which is honest enough for now.

Your dictation history is encrypted. The key lives in the macOS keychain, tied to that machine, syncing nowhere. Transcripts stopped sitting on disk in plain text.

A filter for endings you never said. The model picked up YouTube habits and appended things like thanks for watching. Nobody said them, so they get cut.

Language detection is automatic. Your custom vocabulary can be in any script. The joke inside the project is that an English-first app hears two languages most days.

Models are picked in settings. Both the recognition model and the cleanup one. That used to be a terminal-only job.