Get help with Vocelle

Getting started

Download Vocelle, drag it to Applications, and open it. First run does three things: asks for the two permissions it needs, downloads the speech model, and compiles that model for your Mac's Neural Engine. Each step is covered below. After that, hold right-⌘ anywhere, speak, and release.

Permissions: Accessibility and microphone

Vocelle needs exactly two permissions, and it can't work without either:

• Microphone — so it can hear you. Grant it in System Settings → Privacy & Security → Microphone, and switch Vocelle on.
• Accessibility — so it can detect the right-⌘ hold system-wide and type the result at your cursor. Grant it in System Settings → Privacy & Security → Accessibility, and switch Vocelle on.

If dictation seems to record but nothing types, Accessibility is almost always the missing one. If the waveform overlay never appears when you hold right-⌘, check Accessibility too — key detection needs it.

Still stuck after granting both? Toggle the permission off and on again in System Settings, then quit and reopen Vocelle. macOS occasionally needs the app restarted to notice a fresh grant.

"I don't see the menu-bar icon"

Vocelle lives in the menu bar — a small waveform icon. If you don't see it:

• Menu-bar managers hide it. If you use a menu-bar management tool (Bartender, Ice, Hidden Bar, and similar), Vocelle's icon is very likely tucked into its hidden section. Open your manager's settings and mark Vocelle as always visible.
• A crowded menu bar hides it. On smaller displays, macOS drops icons that don't fit — especially with an app open whose menus are long. Close a few other menu-bar apps or check with a different app in focus.
• Confirm it's actually running. Open Vocelle from Applications again; if it's running, that brings the icon (and a settings window) forward.

The icon being hidden doesn't stop dictation — hold right-⌘ and Vocelle still works. The icon is just the door to settings and history.

First run: the model download and compile

On first launch, Vocelle downloads the Whisper large-v3-turbo speech model — about 1.6 GB, once. This is the only network request the app ever makes. How long it takes depends on your connection.

After the download, macOS compiles the model for your Mac's Neural Engine. This can take a few minutes and happens exactly once; the progress indicator isn't stuck, it's compiling. Subsequent launches are fast, and dictation works fully offline from then on.

If the download is interrupted, reopen Vocelle and it resumes. If it fails repeatedly, check that your network or firewall isn't blocking the download, and email us if it persists.