A UNIVERSAL INTERACTION LAYER FOR YOUR SCREEN
If you can see it,
you can interact with it.
Press a shortcut, point at something on your screen, and RPCTION selects the thing, not a rectangle around it. Text, a link, a phone number, an object in a photo, something moving in a video, code in a screenshot. Then copy it, extract it, or keep it.
It runs entirely on your machine. No account, no cloud, nothing uploaded. RPCTION is currently available for Windows and Linux.
Check this out:
youtube.com/watch?v=dQw4w9WgXcQ
call me when you get a sec, +972 50 123 4567
ship it to 1600 Amphitheatre Parkway, Mountain View, CA 94043
export function total(items) {
let sum = 0;
for (const it of items) sum += it.price;
return sum;
}SIMULATED OVERLAY · MOVE YOUR POINTER OVER THE SCREEN TO TAKE CONTROL SIMULATED OVERLAY · TAP A TARGET CLICK A TARGET
THE PROBLEM
Your screen was never really static.
A screenshot turns a working link into a picture of a link. The phone number stops being callable. The table stops being data. Everything on your display is already a URL, a number, an address, an object, right up until the moment you need to use it.
WITHOUT RPCTION
Retype it. Or squint at it and hope.
WITH RPCTION
Point. Click. It is on your clipboard.
HOW IT THINKS
See
RPCTION reads the pixels your display is already showing. It does not ask the application for anything, and the application never knows it is there.
Understand
Text, links, phone numbers, addresses, QR codes, images, tables, object boundaries, recognised on your machine and arranged into a hierarchy of what is under your cursor.
Act
You get the options that make sense for the thing you pointed at. Copy it, extract it, open it, save it. Then RPCTION disappears.
Replicate
What lands on your clipboard is the thing itself, the URL, the sentence, the object with its background removed. Not a screenshot of it.
WHO IS IN CHARGE
You point. You decide.
RPCTION is not an autonomous agent. It does not act on your behalf, and it does not guess at your intent. It recognises the thing under your cursor and offers the options that fit it.
Copying is instant, because copying is harmless. Anything that leaves RPCTION, opening a browser, a dialer, a map, is marked with an arrow and never happens on a plain click.
NOTHING LEAVES THE DEVICE
MODES · CYCLE WITH TAB
Thirteen ways of looking at the same pixels.
AUTO handles almost everything. The rest are there for when you want to be specific, hover one and watch what it would select.
Check this out:
youtube.com/watch?v=dQw4w9WgXcQ
call me when you get a sec, +972 50 123 4567
ship it to 1600 Amphitheatre Parkway, Mountain View, CA 94043
export function total(items) {
let sum = 0;
for (const it of items) sum += it.price;
return sum;
}PREVIEW FOLLOWS THE SELECTED MODE
OBJECT MODE
Not a crop.
A rectangle around a thing is not the thing. RPCTION works out where the object actually ends and cuts along that boundary, so what you paste is the object on transparency.
When the boundary is genuinely ambiguous it says so rather than inventing one, the outline goes dashed, the confidence drops, and a plain rectangular copy stays one keystroke away.
Check this out:
youtube.com/watch?v=dQw4w9WgXcQ
call me when you get a sec, +972 50 123 4567
ship it to 1600 Amphitheatre Parkway, Mountain View, CA 94043
export function total(items) {
let sum = 0;
for (const it of items) sum += it.price;
return sum;
}DASHED OUTLINE = INFERRED BOUNDARY
TEXT INSIDE AN IMAGE INSIDE A CHAT
TEXT MODE
Text doesn't stop being text because it's a screenshot.
Someone pastes a screenshot of a conversation. The words are right there, and completely useless. RPCTION reads them back off the screen and hands you the sentence, the paragraph, or the single word, whichever you pointed at.
No dragging a rectangle. No hoping you got the edges right.
LINK & CONTACT MODES
A URL is still a URL.
Even when it is trapped inside an image, a screenshot, a video frame, a screen share, a chat bubble, a PDF, or an application that was never going to let you select it.
The same is true of a phone number, an email address, a postal address, and a QR code. RPCTION recognises the kind of thing it is, then offers actions that fit that kind.
| Region | Revenue | Units | Share |
|---|---|---|---|
| North | 412300 | 1820 | 34% |
| South | 288100 | 1240 | 24% |
| East | 355900 | 1602 | 29% |
| West | 151200 | 705 | 13% |
TABLE MODE · STRUCTURE PRESERVED AS TAB-SEPARATED DATA
MOTION MODE
Copy what you see moving.
Everything else works on one frame. Motion works on a thing across time: point at something moving, in a video, a game, a screen share, and RPCTION follows it and takes it with you.
This is not a recording with a crop on it. The thing being copied is the target, not the screen it happened to be on. Choose two, five or ten seconds, then take the region it travelled through, or the entity itself, isolated on transparency.
SOURCE MODE
Get the code back out of the picture.
Code in a screenshot, a slide, a video, a message thread. Point at it and RPCTION reads it back into text you can paste, indentation and all, rebuilt from where the characters actually sit on screen.
It tells you where the text came from, every time. Something rebuilt from pixels is labelled RECONSTRUCTED, not passed off as the original file, and a region it could not read honestly says UNAVAILABLE rather than inventing plausible code.
RPCTION reads what is on your screen. It does not reach into other applications, attach debuggers, or fetch anything over the network, so the two highest-fidelity answers are not available to it yet, and it says so instead of approximating them. Nothing is uploaded.
AUDIO MODE
Keep the last few seconds you just heard.
A line in a call, a phrase in a video, a sound in a game. By the time you decide you want it, it has already gone. AUDIO keeps a short rolling window of what your machine is playing, so you can take the moment after it happens.
Nothing is buffered until you cycle to AUDIO, and the buffer is released the moment you leave. RPCTION is not recording in the background, it cannot be, because the buffer does not exist until you ask for it.
Windows mixes system audio before applications can see it, so AUDIO captures everything you are hearing rather than one application on its own, and says so rather than promising isolation it cannot deliver. Audio stays on your device. There is no transcription service, no recognition service and no upload.
APPLICATION AGNOSTIC
It doesn't care where it came from.
RPCTION works from what your display is drawing, so there is nothing to integrate and nothing to install into anything else. If it is on screen, it is in scope.
Named as examples of the kind of thing you might be looking at. RPCTION has no integration with, and no affiliation to, any of them.
LOCAL PROCESSING
Your screen
stays your screen.
The recognition engine, the model it uses, the QR decoder and the object segmenter are all bundled with the application and run on your machine. Pull the network cable and everything above still works.
The only things that ever leave RPCTION are the actions you explicitly pick, OPEN, SEARCH, MAP, CALL, MESSAGE, which hand the selected value to another application on your own computer.
DESIGN PRINCIPLE
Invisibly visible.
RPCTION does not want to be another window on your desktop. There is no panel to arrange, no sidebar to collapse, nothing docked anywhere.
It appears exactly where you are pointing, draws a single outline and a two-word label, and then gets out of the way. Most of the time you cannot tell it is running at all.
SPEED
Point.
Click.
Done.
Recognition runs on the region under your cursor first, so the thing you are actually pointing at becomes usable while the rest of the screen is still being read. No spinner. No progress bar. No waiting for a server.
About 30 ms median hover analysis in our test environment. Your machine is not our machine, what the number is there to say is that the work happens between you moving the pointer and you noticing, because it happens on your own hardware rather than a round trip to somewhere else.
WHEN IT IS NOT SURE
Wrong is worse than empty.
RPCTION has no object classifier and does not pretend to certainty it has not got. Some scenes are genuinely ambiguous, two objects of the same colour that touch, something behind glass, a heavily compressed photograph, and in those it can pick the wrong thing.
So it shows you first. The selection is drawn while you hover and acting on it takes a click, with ↑ ↓ to walk out to the container or back in to the part. A wrong answer you can see is survivable. A wrong answer you cannot see is not.
GET RPCTION
Your screen.
Your selection.
NO SIGNUP · NO SUBSCRIPTION · NO CLOUD
Version 0.3.0 · Windows and Linux · 64-bit · free · not code-signed.
No account. No subscription. No cloud.