ChatGPT on a Stream Deck: Keys, Icons and Voice

The Voice Pad is a compact mechanical keypad supplied with a microphone: three keys marked for microphone, tick and cross, plus one knob, $79. It is the shape most people are really after when they search for ChatGPT on a stream deck. One key to start talking, one to accept what came back, one to throw it away.

What do people actually map for ChatGPT?

Four things come up over and over: bring the application to the front, start dictation, paste a standing prompt you retype constantly, and accept or discard a suggestion in the editor. All four are keystrokes underneath. Anything that can send a keystroke can do all four, which is why this turns out to be a question about labels and feel rather than about capability.

The prompt one is the sleeper. A key that types a full paragraph of instructions you would otherwise retype from memory saves more time per day than the accept key does, and it costs one slot.

Do I need a plugin for this?

No, and the boring route is the sturdier one. Give the application a shortcut at the operating system level, then map a key on the pad to that same chord. The pad is now pressing the shortcut you would have pressed yourself, and nothing sits in between waiting to break on an update.

A deck with screens does have a software library where other people publish integrations for particular applications, and that can save you the setup. We do not build for that library and cannot tell you what is in it this month, so we are not going to pretend to know.

What about the icon on the key?

This is where the screen version genuinely wins and it is worth saying so plainly. If a key has a display under it, the icon is an image you assign, and you can change it the next time a tool changes its logo. A mechanical key has no display, so the label is the cap itself: printed, or blank and left to memory.

In practice people stop reading labels after roughly a week and go by position instead, which is why blank caps worry buyers a great deal more before purchase than after it. If you want the marks on your desk for their own sake rather than on a working key, the Keycap Fidget at $19 is three keycaps on a clear base, each printed with a different AI tool's mark. It is a desk object. It plugs into nothing and presses nothing for you.

Is a microphone key better than a key that types?

It depends entirely on whether you are willing to speak. Talking a paragraph into a chat model is faster than typing it, and a push to talk key is what makes that bearable, because the microphone is only live while the key is held rather than listening all day. The Voice Pad at $79 puts that on a dedicated key with the tick and cross beside it. The Whisper Key at $87.90 is the handheld version, a clear polycarbonate body you hold up to your mouth and speak into.

In an open-plan office the honest answer is often that you will not use either of them, whatever you tell yourself in the checkout. If that sounds like your desk, buy the key that pastes text and skip the microphone entirely.

For the terminal side of the same habit, see using voice input with terminal coding agents.

Back to blog