What system-wide dictation should do
A system-wide dictation app follows the active cursor instead of asking you to compose inside a separate window. The basic loop is simple: focus a text field, hold a shortcut, speak, release, and receive text in that field.
That makes setup and recovery behavior important. You should know which shortcut starts listening, when recording stops, where text will appear, and whether you can review rewritten text before it changes your work.
- Work in ordinary Windows applications and browser fields
- Make listening state obvious
- Keep shortcut behavior predictable
- Show whether output is verbatim or rewritten
Dictation and cleanup are different jobs
Speech recognition turns audio into words. Cleanup decides whether those words should be rewritten. Treating those as separate choices prevents a common surprise: asking for an exact transcript and receiving a polished sentence instead.
Quill exposes two shortcuts. Dictation types every spoken word, including self-corrections. Scribe resolves the correction first and presents the cleaned wording for review. The shortcut you press declares which behavior you want before recording begins.
A practical first test
Try the same sentence in an email, a browser field, and a notes app. Include a name, punctuation, and a spoken correction such as “Tuesday—no, Thursday.” Check whether the result lands at the cursor and whether the correction behavior matches the mode you selected.
Quill currently treats Windows 10 and 11 as the verified path. It can run recognition locally and can optionally use a cloud provider when you explicitly choose that route.