It does not just tell you where to click.
It points.
Hold a key and ask out loud. The companion looks at what is actually on your screen, talks you through it, and moves the cursor to the exact control while it is still mid-sentence. In any app, or in the browser with nothing to install.
Open the Export tab up here.
1/4Watch the cursor
The pointer moves before the sentence ends.
Most assistants finish thinking, then answer. This one streams: the moment it decides which control you need, the cursor is already on its way, so the words and the movement arrive together the way they would if a person were sitting next to you.
It circles, underlines, and draws arrows between things when that explains it faster than talking does.
Why it lands on target
It reads the interface, it does not squint at a screenshot.
The usual way
Guess a pixel from an image
Take a picture of the screen, shrink it, ask a model for coordinates, then scale those back up. Every one of those conversions loses a little, the model adds its own aim error, and the miss gets worse on a second monitor or a scaled display.
How the companion does it
Ask the interface where things are
Every app already publishes its controls and their exact positions, because that is how screen readers work. The companion reads that list, so it points at the real button rather than a guessed spot. No rescaling, no drift, and it stays correct when you scroll or resize.
When an app publishes nothing useful, which happens with some canvas and game interfaces, it falls back to reading the screen. In that case the pointer is drawn more faintly on purpose: a guess should not look as confident as a certainty.
Everywhere you work
One companion, three places, no second-class version.
macOS
DownloadA menu bar app. Hold the hotkey anywhere, in any application, and it sees what you see.
Get itWindows
DownloadThe same app from the same codebase, not a port and not a waitlist. Most rivals are Mac only.
Get itIn the browser
Nothing to installThe widget and the extension run the same companion on any web page, with no download at all.
Open the demoThe desktop app and the browser version share one engine, so a fix in either is a fix in both, and what you learn in one works exactly the same in the other.
Questions
The things people actually ask.
- How is this different from asking a chatbot?
- A chatbot describes a menu you then have to go find. The companion moves your cursor to it while it is still talking. You are never translating instructions into clicks yourself.
- Does it click things for me?
- No. It points, circles, and explains. Nothing is clicked without you, and it never synthesises button presses an app could mistake for real input. You stay in control of your own machine.
- What does it send anywhere?
- The list of controls it can see and your question. In the browser, speech recognition and the voice are handled by your own device with no audio leaving it. Bring your own key and the only server involved is the model provider.
- What does it need on my machine?
- On Windows, nothing: no permission prompts at all. On macOS it asks for Screen Recording and Accessibility, and macOS requires a relaunch before either takes effect. Accessibility is what makes the pointing exact rather than approximate.
- Does it work if I cannot see the screen well, or use a screen reader?
- That is the origin of the whole product. Everything it says is spoken, every step is narrated in plain language, and the on-page version respects reduced motion and works from the keyboard.
Stop hunting for the button.
Try it on any page right now. It takes about ten seconds and needs nothing installed.