FAQ

A universal visual interaction layer for your screen. You point at something you can see, and RPCTION works out what it is and what you can do with it.

No. A screenshot tool captures a rectangle. RPCTION identifies the thing you are pointing at, the URL, the sentence, the object, and gives you that thing rather than a picture of the area around it.

Not in the sense people usually mean. There is no chatbot, no language model, and nothing is sent to a service. Recognition uses a bundled text-recognition engine and deterministic local algorithms for patterns, QR codes, image regions, object boundaries and tracking. It all runs on your machine.

No. Screen contents are processed locally and are never transmitted. The only things that go outward are actions you explicitly choose, like opening a URL in your browser.

Yes. After installation the recognition engine needs no network at all.

Neither.

RPCTION works from what your display is drawing, so it does not need an integration with any application. If the content is visible on screen, it is in scope. It has no affiliation with any of those products.

Yes, that is one of the things it is best at. Text inside an image is still text as far as RPCTION is concerned.

Windows 10 and 11. macOS and Linux are not supported. The architecture is built to extend to them, but nothing has been tested and nothing is being claimed.

Not yet. Version 0.1.0 installs on Windows 10 and 11.

It is a local, edge-aware segmenter, not a learned model. It is genuinely good on logos, icons, product shots and objects that separate from their background. On cluttered scenes it reports low confidence and offers a rectangular copy instead of inventing a boundary.

Point at something moving, click to lock, and RPCTION follows it. Then take 2, 5 or 10 seconds of it, either the region it travelled through, or the entity isolated on a transparent background.

No. It keeps a bounded rolling buffer of recent frames in memory, only while you are in Motion Mode, and discards it when you leave. Nothing is written to disk unless you choose to save an export.

Text, and things. Paragraphs, lines and single words; URLs, email addresses, phone numbers, addresses, tables and QR codes; images, and objects inside images, a person, an animal, a car, a product, a logo, an icon. Also a component of an object rather than the whole thing, or a surface rather than the object wearing it.

Yes, that is the point of the tool. OBJECT mode selects the object rather than a box around it, and extracts it onto transparency. It has no object classifier and never names what it found: it works out where the thing ENDS, structurally, which is a different question from what the thing IS.

A mode that reads code off your screen and rebuilds it locally, from an editor, a screenshot, a slide, a paused video, restoring the indentation from where the characters actually sit. Every result states where it came from, and offers COPY or SAVE with an extension matching what the code actually is.

No. RPCTION reads pixels. It does not attach to other applications, read a browser’s internals, use a debugging port, install an extension, or fetch the page over the network, so private, server-side and unloaded code is out of reach, and no amount of pointing at a page will produce it. What SOURCE gives you is the code that is VISIBLE, reconstructed, and labelled as reconstructed. Nothing in this release can report SOURCE AVAILABLE, which is reserved for a real source file obtained legitimately.

A mode that keeps the last few seconds of what your machine is playing. Cycle to AUDIO, choose 2, 5, 10 or 30 seconds, click, and it saves a .wav. Useful for the line in a call or the phrase in a video that has already gone by the time you decide you want it.

No, and it could not be: the buffer does not exist until you cycle to AUDIO. Entering the mode allocates it, leaving releases it, and it has a fixed maximum size so it cannot grow into a recording. With RPCTION idle, Windows audio modules are not even loaded into the process.

No. Windows mixes system audio before applications can see it, so AUDIO captures everything you are hearing. We would rather say that than promise isolation the platform cannot deliver.

No. ACT is user-directed at every step: you select the target, RPCTION identifies it, you choose the action, and only then does anything happen. It is not an agent, it does not browse, and it will not purchase, send, post, delete, submit or confirm anything without you choosing that action.

Only through your own browser, and only when you ask. Choosing OPEN hands a URL to your browser; MAP hands an address to a maps service. Those are the outbound paths, they are marked in the interface, and none of them happens on a plain click.

You see it before anything happens, the selection is drawn while you hover, and acting takes a click. Up and down arrows walk the hierarchy out to the container and back in to the specific thing. Some scenes are genuinely ambiguous to RPCTION and it will sometimes pick wrongly; the design principle is that a wrong answer you can see is survivable and a wrong answer you cannot see is not.

None.