A Keyboard-First Dictation Workflow

The failure mode for a dictation tool is not bad accuracy. It is friction. If starting a dictation means finding a window, clicking into it, talking, selecting the result, copying it, and switching back, you will use the tool for the first week and then go back to typing.

So the design question for Voice was not "how good can the transcript be." It was "how few things can happen between the thought and the text." The answer we landed on is keyboard-first, and every other decision follows from that.

No window to manage

Voice lives in the menu bar. Press the shortcut, which is ⌥⇧Space by default, and a panel opens right under your cursor in whatever app you are already in. Press it again to stop, and again to copy. There is no application window to find, no dock icon to hunt for, and no context switch out of the document you were writing.

That matters more than it sounds like it should. The cost of a context switch is not the two seconds of clicking. It is that you arrive in the other window and have to reload what you were about to say. Dictation is useful precisely when you have a sentence ready in your head, and a window switch is very good at deleting that sentence.

Push to talk versus toggle

These are two genuinely different working styles, and neither is correct for everyone, so Voice has a shortcut for each.

• Toggle: press once to start, press again to stop. Your hands come off the keys while you talk. This suits longer dictation, where you want to gesture, pace, or look at something else while you speak.
• Push to talk: hold to record, release to stop. It suits short bursts, a sentence at a time, and it makes the end of the recording unambiguous. You physically cannot forget to stop it.

The reason to support both is that toggle has one specific failure and push to talk has another. With toggle, you walk away from a live microphone because you thought you pressed stop. With push to talk, a long dictation means holding a key for ninety seconds, which is genuinely uncomfortable. People sort themselves into these camps quickly. The tool should not pick for them.

The shortcuts that actually earn their keys

Every action in Voice has a shortcut you set yourself in Settings, and it works from any app. The set is deliberately small:

• Dictate: opens at the cursor and starts recording, defaulting to ⌥⇧Space.
• Push to talk: hold to record, release to stop.
• Show or hide the panel: opens and closes without recording, for when you want to look at the last result.
• Copy last transcript: copies the previous result without opening anything, which is the one people discover late and then use constantly.
• Cancel dictation: stops and discards the current dictation from any app.

Cancel deserves a note. It looks like a minor convenience and it is actually what makes the rest comfortable. If the only way out of a dictation is to finish it, every accidental trigger becomes a small chore, and you start hesitating before you press the key. A guaranteed escape hatch, bound to a key you can hit without looking, is what lets you start a dictation casually.

Why the bindings have to be configurable

A global shortcut is a claim on a key combination across the entire system, and the system is crowded. Your editor, your terminal, your window manager, your screenshot tool, and whatever your company installed all have their own claims. Any default a tool ships will collide with something on someone's machine, and when it does, the collision is usually silent. One app gets the keystroke, the other does not, and you are left thinking the app is broken.

So in Voice you click a shortcut in Settings and press the keys you want. You can clear one with ⌫ if you would rather not have that action bound at all, or restore the defaults if you have made a mess of it. This is not a power-user feature. It is the difference between a tool that works on your machine and one that works on ours.

Review before the clipboard, and never into the document

This is the decision people push back on most, so here is the reasoning.

Voice does not type into your apps. Nothing reaches your clipboard until you have looked at the text and chosen Copy. Then you paste it yourself with ⌘V. An app that types directly into whatever window has focus saves you exactly one keystroke, and takes on several problems in exchange.

It has to be right about which window has focus, at the moment the text is ready rather than the moment you started talking. If you tabbed away to check something while you spoke, the text lands somewhere you did not intend. It has to synthesize keystrokes, which requires broad accessibility access to your machine and interacts unpredictably with apps that do their own input handling, particularly editors with autocomplete and terminals with their own key interpretation. It makes a mess of undo, because your document now has a change in its history that you did not make and cannot always reverse cleanly. And it commits before you have read anything, which means the first time you see a misheard name is after it is already in the document.

The clipboard avoids all of that by being a handoff rather than an action. The text sits there until you decide where it goes. You keep the paste, which means you keep the ability to put it in a different field, paste it twice, or not paste it at all.

Reviewing is faster than correcting

The panel shows the cleaned text with the original one toggle away, a word count, and Discard, Redo, and Copy sitting together. Copy is bound to Return, so the common path is press shortcut, talk, press shortcut, glance, press Return.

That glance is the point. Reading a short block of text takes a couple of seconds. Finding a wrong word after it has been pasted into a message means rereading, clicking into position, selecting, and retyping, and that is the good case. The bad case is that you do not find it. Putting the review before the commit is not caution for its own sake. It is the cheaper order of operations.

The takeaway

Keyboard-first is not about saving keystrokes. It is about keeping a tool close enough to reach without a decision, then making it stop short of the thing you should decide yourself. A shortcut for every action, bindings you choose, an escape hatch that always works, and a deliberate pause before anything touches your document. Tools that skip that last part feel faster for a week and then cost you an afternoon.

Voice runs on Apple Silicon Macs with macOS 15 or later, and we distribute it directly rather than through a public installer. If you want access, tell us what you would use it for.