Your voice is not a data source.
Voice is unusually personal input. Cloud transcription sends that input to a remote service, where privacy depends on someone else’s retention policy, security, and business model. Saylight takes the opposite route.
Speech recognition runs entirely on the Mac using local models. Recordings are not uploaded for transcription, and there is no account or cloud service behind the writing. Dictation keeps working offline because the tool does not need to phone home with your voice.
Private without becoming inconvenient.
Place your cursor in an email, message, or note. Hold the fn/Globe key, speak, and release. Saylight inserts the locally transcribed words where you were already writing. A hands-free mode, speech buffer, searchable history, and scratchpad stay close without turning dictation into another online destination.
Audio is never saved to disk. Temporary speech-buffer recordings remain in memory only while they are useful. Completed transcripts and notes can be saved locally with owner-only permissions; history can be paused, cleared, or disabled without giving up dictation.
No cloud transcription.
Bundled models turn speech into text on the Mac. Recordings and dictated text are not sent to a transcription provider.
The recording disappears.
Audio stays in memory while it is needed and is never saved to disk. Clearing a rehearsal or quitting releases temporary takes.
Your words, under your control.
Saved transcripts and notes remain in private local files. Pause history, delete individual entries, or clear it entirely.
Names stay private too.
Dictionary terms, context for corrections, and recurring mishearings stay on the Mac and are never transmitted.
Privacy means being specific.
Saylight needs Microphone permission to record when asked and Accessibility permission to insert text into another app. It uses a short-lived clipboard entry for broad compatibility, then restores the previous clipboard when it has not changed. It does not silently record in the background.
The software is an early release for macOS 13 or later, with downloads for Apple Silicon and Intel Macs. It is an example of what useful AI can look like without data collection as the price of admission: a familiar shortcut, a local model, and your own words on the page.