“Voice control on a Mac” means at least three different things, and most of the frustration people have with it comes from wanting one of them and being handed another. This page separates them, says what macOS already does for free, and is honest about where each approach stops.
The three things people mean
Before choosing a tool, work out which of these you actually want. They are solved by different software and the good answer to one is a bad answer to the others.
| You want | The category | Example of the ask |
|---|---|---|
| Text, typed for you | Dictation | “Write this paragraph into the email.” |
| The Mac operated, click by click | Command and control | “Click Send.” |
| An outcome, not the steps | A voice agent | “File these into the right folders.” |
1. Dictation: already on your Mac, and good
macOS has system-wide dictation built in. Turn it on in System Settings › Keyboard › Dictation, then press the shortcut (fn twice by default) with your cursor in any text field. On Apple Silicon many languages run on-device, so it keeps working with no network and nothing leaves the machine.
It is genuinely good and most people do not know it is there. If all you want is your words typed into a box, you may be finished after this section — there is a fuller comparison of the dictation tools on dictation on a Mac.
Where it stops. It transcribes literally. It does not know what is on your screen, so it cannot answer “reply that Tuesday works” — it will type the words “reply that Tuesday works”. Punctuation has to be spoken. And it does nothing outside a text field.
2. Voice Control: the most underrated feature in macOS
Apple's Voice Control lives in System Settings › Accessibility › Voice Control. It is free, it works offline after a one-time download, and it will drive the entire machine.
- Say a button's name and it clicks it: “Click Send”.
- Say “Show numbers” and every clickable element gets a number you can say instead — this is the trick that makes it work in apps with unlabelled controls.
- Say “Show grid” to divide the screen and drill down to any point.
- You can define custom commands that run a sequence of keystrokes.
If you need hands-free operation of a Mac for accessibility reasons, this is the tool, and nothing described further down replaces it. It is precise, it is local, it is free, and it is supported by Apple.
Where it stops. Every command has to be one you can name. Voice Control is a faithful translator from speech to clicks, which means you still supply the plan. A ten-step task is ten commands, in order, correctly remembered. It also cannot answer a question: ask it what a setting does and it will look for a button called “what a setting does”.
3. Siri and Shortcuts: automation, if the app cooperated
Siri can run Shortcuts, and the Shortcuts app can chain real actions together. When an app has shipped the right intents this is excellent and near-instant.
Where it stops. On the words “when an app has shipped the right intents”. Coverage is decided by each developer, so it is patchy and it is worst in exactly the places most work happens: internal tools, admin panels, web apps, and anything niche. If nobody wrote an intent for your task, there is no route to it.
What a voice agent adds
The newer category takes the goal rather than the command. You describe the outcome; the software reads the screen, decides the steps, and carries them out by moving the pointer and pressing keys — the same inputs a person produces, so it needs no cooperation from the app.
Two capabilities distinguish this from everything above:
- It can answer questions about your screen. “What does this setting do?”, “Where is the export button?” Dictation and Voice Control cannot do this at all.
- It supplies the plan. One sentence can stand for a sequence you did not have to work out.
The honest caveats
Anyone selling you this category without the following list is selling you something.
- It is probabilistic. It will sometimes do the wrong thing. Any tool that takes your pointer needs a stop key you can reach without the pointer, and a way back from an action that already happened.
- It costs money to run. Frontier models are not free, so this category is priced per use in a way the built-in features are not.
- It is slower than a keyboard shortcut. If you already know the shortcut, use the shortcut. A good agent should be reaching for the shortcut itself rather than clicking through a menu on your behalf.
- It needs deep permissions. Reading the interface means Accessibility access; seeing the screen means Screen Recording. Be suspicious of anything asking for more than it needs — Full Disk Access, for example, is not required to read the window in front of you.
How to choose
| If you want | Use |
|---|---|
| Words typed into a field | Built-in Dictation. Free, on-device, already installed. |
| Hands-free control of everything | Voice Control. Free, offline, precise, Apple-supported. |
| Automation of a task you do daily | Shortcuts, if the apps involved support it. |
| Answers about what is on screen | A voice agent. Nothing built in does this. |
| An outcome without naming the steps | A voice agent. |
The first three are free and are the right answer more often than a vendor will tell you. Try them before paying for anything.
Where Reflick fits
Reflick is a voice agent for macOS, and it is the product this site sells, so treat this section as the interested party talking.
It lives in the menu bar and the notch. Hold ⌃⌥ and ask about your screen; hold ⌥⇧ to dictate; say “reflick agent” and it performs the task itself. Three design decisions follow from the caveats above:
- It tries the free answer first. Seven rungs, cheapest first, and five of them never touch a network: the control under your pointer read from the Accessibility tree, device state read from the system, 83 keyboard shortcuts it already knows, and 112 native macOS capabilities. A model is the last resort. Talking is free and unlimited on every plan, including the free one; only a task Reflick performs for you is metered.
- Estimated coordinates may draw, never click. Clicks are aimed at real Accessibility frames. A traced line twenty points off is still readable; a click twenty points off is the wrong button. Escape stops a run, and it is a key rather than an on-screen button because an agent driving your Mac must not be able to click its own stop control.
- Three permissions and no more. Accessibility, Screen Recording, Microphone. Full Disk Access is never requested. There are no OAuth connections and no account tokens, because Reflick talks to the apps already running on your Mac rather than to a cloud API on your behalf.
Reflick 1.0 is out and you can download it for Mac now — a signed, notarised app, free to start with 30 agent runs a month and unlimited talking. There is a longer list of questions if you want the detail, a separate guide to dictation on a Mac, and every Mac voice tool compared.