Permissions

Screen & System Audio Recording

Listing applications and capturing their audio require Screen & System Audio Recording permission, which macOS asks for the first time ScribeKit looks for sources. Without it, the setup screen reports the missing permission instead of a list or a capture.

ScribeKit does not pre-check the permission in order to avoid asking for it: the first discovery attempt is what makes macOS ask, exactly as the normal permission flow expects. When an attempt fails, ScribeKit asks macOS through CGPreflightScreenCaptureAccess whether the access exists, so it can say that access is missing rather than repeat a framework error. That check never prompts, and nothing polls for the permission to change.

Because a permission that has never been asked for and one that was refused look the same from here, ScribeKit says the access is not available rather than claiming you denied it. Grant it in System Settings › Privacy & Security › Screen & System Audio Recording, then choose Refresh on the setup screen. ScribeKit does not claim a relaunch is needed.

ScreenCaptureKit has no audio-only stream, so the capture filter names a display as well as the selected applications, and the permission macOS asks for is therefore the screen recording one. No screen output is added to the stream, so no frame is delivered and no video or screen content is ever processed. See Capturing App Audio.

Speech recognition: none

Transcription asks for no permission at all, because it runs against a speech model on this Mac rather than through a service.

What it does need is the model itself. ScribeKit lists the languages the recogniser supports, marks the ones whose model is not installed, and refuses to start in a language it cannot transcribe — it never downloads a model, never substitutes another language, and never falls back to network recognition. Use Check Again on the setup screen after installing one. A Mac that cannot run the recogniser at all is reported as that, rather than as a list with no languages in it. See On-device Speech.

Files: the folder you picked

ScribeKit reaches the filesystem only through a location you chose in a system panel, persisted as a security-scoped bookmark. Access is started for the work that needs it and stopped afterwards. The App Sandbox is never weakened or disabled to reach a path, and ScribeKit never writes to a location you did not pick. See Save Location.

Network: no entitlement

The app ships without the network client entitlement, so the sandbox does not permit it to open a network connection at all. See Network Policy.

What a real Mac reported

On a development Mac that had already granted Screen & System Audio Recording, had an installed en-US speech model and had a save folder remembered from an earlier launch, the running app reported all four prerequisites as Ready before any application had been selected, and the process held no network sockets at all while it ran.

That is the satisfied path, observed. Refusal has still never been reproduced on a real machine: no permission has been denied, no disk pulled and no speech model uninstalled to see what ScribeKit says. Those states are reached through injected failures in the test suite, and are listed in Limitations.