Dictate

Push-to-talk dictation into any app, with speech models that run on your computer.

Dictate types what you say into the app you're working in. Hold the shortcut, speak, and let go: the words land where your cursor was. Speech is recognized on your computer by a model you download once.

Set it up

The setup's Dictate step and Settings › Dictate › Set up walk through the same choices (the toggle shortcut and the backup microphone are rows on Settings › Dictate itself):

• Shortcut. Ctrl+Alt on Linux. Hold it in any app to dictate; double-tap it to keep listening hands-free, and press it again to stop. A quick tap does nothing, and holding Ctrl+Alt while you press another key (Ctrl+Alt+T) is left to that shortcut. An optional toggle shortcut (Settings › Dictate) starts and stops a take with single presses. Esc cancels a take, however it started.
• Try it. Press and release your shortcut once, so Vibezeo knows the key reaches it.
• Microphone. Pick one, or follow the system default. Settings › Dictate adds an optional backup, used when the first one is unplugged, or delivers no sound in the first second of a take even after Vibezeo reopens it; there's no backup unless you pick one. A Bluetooth headset's mic is never picked on its own while another mic exists, which keeps headphones out of low-quality call mode, and neither are loopback or virtual inputs.
• Model. Download a speech model (below).
• Where text goes. Into the focused app (the default) or onto the clipboard. After a paste, the clipboard gets back what it held before.

Models

Models are downloaded only when you ask, from Hugging Face at pinned versions. Each file is checked against a pinned SHA-256 and size before it's used, and a file that doesn't match is deleted.

Manage models… in Settings › Dictate downloads, switches and removes them; on Linux it marks Distil-Whisper as CPU transcription. A switch during a take applies when the take ends.

While you talk

The widget shows Listening with a timer, then Transcribing, then how many words went in. It appears while recording (or always, or never: Show widget; a take started from the tandem://dictate/toggle link shows it anyway, so you can see one is running), and its × discards the take. A take stops at 10 minutes; the widget counts down the last 30 seconds. Takes too short or silent to transcribe insert nothing and say why.

In Vibezeo's own message boxes, the mic button records while you hold it or toggles with a click, and so does holding your Dictate key while the box has focus; the text goes in at the caret without touching the clipboard. The status bar shows your Dictate key, and the tray menu shows Dictate's status.

History, stats and the dictionary

• History keeps the text of each take on this computer, never the audio: forever by default, or 90, 30 or 7 days. Search, copy, correct, delete, export to CSV, or clear it; clearing keeps your totals.
• Stats count words, speaking time, takes and words per minute, and estimate time saved against typing at 40 words a minute.
• Dictionary holds words Dictate should spell your way and corrections ("tandem" → "Tandem"), optionally only in some apps or languages, and a rule can leave links, email addresses and code alone. Its Review tab suggests names and terms that keep coming back; only Remember adds them. Deleted rules can be restored for 30 days, and the dictionary imports and exports JSON or CSV.
• Sync dictionary (Comes with Vibezeo accounts. It uses Vibezeo's servers and switches on when Vibezeo accounts open. Off until you turn it on) keeps the same rules on each computer you sign in to. When the same rule changed on two computers, the newer change wins, and an older one you made offline asks which to keep. Turning it off keeps your rules here; the synced copy stays in your account until you choose Stop and delete synced copy.

Linux

• X11: nothing else to install.
• Wayland: the shortcut uses your desktop's global-shortcuts portal. When there's no portal, or the shortcut can't be registered or stops working, Settings shows a command (xdg-open tandem://dictate/toggle) to bind to a shortcut in your system settings; the link works only while that's the case. Press it once to finish setup. Typing into other apps needs wtype and wl-clipboard, and Settings shows the install command if they're missing. Esc can't cancel a take there; the widget's × does.

Privacy

With local models, audio never leaves your computer and is never saved; the only network request Dictate makes is the model download you start. Transcribe with › Cloud sends a take to Vibezeo's speech-to-text service instead. It uses Vibezeo's servers, switches on when Vibezeo accounts open, and shows what it sends the first time you switch to it. Share dictionary hints with cloud (off by default) also sends your preferred spellings, up to 40, with each take.

Sync dictionary (off by default) sends your dictionary rules (the words, replacements and their options, and which rules you deleted) to your Vibezeo account. Transcripts, history, audio and the Review tab's examples never sync.

Enhance and Polish send the transcript's text, never audio, to Vibezeo's AI service, even when speech is recognized locally. They're refused over 5,000 characters, and on any failure your exact words are used. Formatting instructions (up to 2,000 characters, and per app) apply to every take as a Polish.

Cloud, Enhance and Polish: Comes with Vibezeo accounts. It uses Vibezeo's servers and switches on when Vibezeo accounts open. Until then Dictate is local and sends nothing.