One shortcut, in any app on your PC. Cursorly works out what's on the screen in front of you and helps at whatever level you asked for, from showing you where to click to running the task itself.
Every mode past pointing is a switch you turn on yourself.
Same shortcut, same reading of the screen. The only thing that changes is how far Cursorly goes once it knows what you meant.
You stay on the keyboard and mouse. A second cursor lands on the exact control you need next, waits for you, then moves again as the screen changes. When a field needs text, it shows you the text.
Works in whatever is open, including software nobody wrote a tutorial for.
Some screens don't need a route, they need a translation. Ask what a control does, or why an option is greyed out, and the answer arrives on the thing itself rather than in a chat window you'd have to map back to the screen.
We're building this now.
Some days you don't want to learn the app, you want the thing finished. Hand the task over and Cursorly carries it out, clicks and all, and it can keep going on a task while you get on with something else.
We're building this now.
Pointing is the resting state and it never goes away. All three modes, in detail.
Any app, any software, anything on your PC.
Sound on.
No second window to learn. It runs inside the app you're already stuck in.
Ctrl + Win + Alt + K out of the box. It fires wherever you are, including the desktop and the Start menu.
Plain words, the way you would say it to a person who knows the software. "Set the font to 14," or "get this into a PDF." There is no command syntax to memorise.
It takes in what's on screen without ever photographing it, then points at the first thing you need. When the screen moves under it, it looks again.
The longer version lives at how it works.
The useful question is not what a tool like this can do on a screen it was demoed on. It's what happens on yours.
The longer version: where it works.
The full version: the privacy policy. Using your own key: bring your own key.
Point. Press the shortcut, type the goal, and a second cursor shows you the next control. Ask and Do are in build now and land as switches you turn on.
Not in the build you can download today, which has no code in it that can press a key or move your mouse. That changes when Do ships, and only if you switch it on.
Usually yes, and that's the point. An in-house tool from 2014 is no harder for it than Word, because it isn't relying on having seen the app before.
Yes. The taskbar, Start menu and Settings are all readable, so you can ask for things outside the app you're standing in.
No. No picture of your screen is captured at any point, so none is ever sent or stored.
A check that runs before anything is read. Password managers, banking, checkout and payment windows end the run. The list is published on what it won't do.
No. It reads when you press the shortcut and stops when the task ends.
No. The planning runs on a cloud model, so it needs a connection.
Signed in, your goal text and task metadata sync to your account. Screen content is never stored or synced.
Yes. Anthropic, OpenAI, OpenRouter or Gemini. The key is encrypted on your machine and you pay your provider directly for what you use.
No pricing yet. This page will say so before anyone is charged.
13 or older.
These have their own page: the FAQ.
Windows 10 and 11. We signed the build recently, so SmartScreen may still warn you the first time. Choose More info, then Run anyway. What the installer does.
It's early, and it gets things wrong sometimes. Something broken, or something in mind? Tell us or book a coffee chat.
Get an email when a new version ships.
It gets things wrong sometimes. Tell us when it does.
Book a coffee chatFifteen or thirty minutes, and there are no slides.
One line is enough.
