SECURITY Signal 142
Open-source offline dictation app FnScribe launched for macOS
FnScribe provides a privacy-focused, offline dictation tool that inserts transcriptions directly into any macOS app without sending data off-device
Because speech recognition runs entirely on the Mac using a bundled Whisper model, no audio or text leaves the device, addressing privacy concerns for voice input. The app’s open-source license lets engineers examine, modify, and redistribute the code, while local dictionary storage ensures user-specific corrections stay on the machine.
Written by elseif from the cluster below · every claim links back to a sourceThe three things worth knowing
FnScribe transcribes speech locally with a bundled Whisper model and inserts the result via a temporary clipboard that restores the previous content.
It requires microphone and accessibility permissions, stores personal dictionary entries in a local settings file with owner-only file permissions, and never transmits them.
Source code is available for inspection and modification, built with Rust and Tauri, enabling macOS users to build or adapt the app.
THE READ
What the cluster adds up to.
FnScribe introduces an open-source dictation utility that runs Whisper locally on macOS and inserts transcriptions directly into the foreground application. The tool lives in the menu bar and activates via a push-to-talk shortcut or hands-free mode. Audio never leaves the device because processing and storage are confined to memory.
Adopting FnScribe requires granting the app microphone and accessibility permissions in System Settings and placing the binary in the Applications folder. Users must select a microphone input and optionally enable launch at login. The bundled Whisper model consumes CPU and RAM during transcription, which may affect performance on older Macs. Recordings are limited to two minutes unless the user switches to hands-free mode, after which they can continue speaking until they stop the session.
FnScribe stops working when the accessibility permission is revoked or when the target app blocks synthetic keyboard input, preventing text insertion. It only transcribes English speech, so non-English users receive no output. Because the source is released under an open-source license, engineers can examine the Whisper integration and propose improvements, but the bundled model itself is not updated automatically.
Written by elseif from the cluster below · checked for specifics the sources never containedTHE CLUSTER