ScreenForm: Your screen is not a picture.
Works over<br>VS Code<br>Safari & Chrome<br>Numbers & Sheets<br>Slack & Mail<br>PDFs & Preview<br>Figma
01<br>Gesture
Three moves, and you never leave the app.
No screenshot, no crop, no pasting context into a chat window. The answer comes to where you already are.
⌃⌘ HELD
⌃+⌘
Overlay on
Hold both keys for a beat and a translucent layer drops over every display. The short delay is why ordinary chords like ⌃⌘F never set it off.
STROKE CLOSED
Region captured
Lasso a chart, an error, a table, a paragraph. Release and the loop closes, everything outside dims, and the edge starts marching.
RETURN
Ask anything about this selection...⌘↑
Answer in place
Pick an action or type your own, then hit Return. Hold ⌃⌘ again to start a fresh selection without closing this one. Esc dismisses everything.
02<br>Input
The circle is a pointer, not the input.
Anything can read pixels. What makes an answer useful is knowing where those pixels came from. So the moment you start drawing, ScreenForm is already gathering the situation around your selection, in parallel and on your Mac, and you never have to explain any of it.
Active app Safari<br>READ
On-screen text on-device OCR<br>READ
Browser tab URL + element<br>READ
Open file VS Code, Cursor<br>READ
This is what makes it useful
Circle a diagram in an article and ask what it means. The pixels are four boxes and some arrows. The page is what says the subject is home heating, so the loop is running in reverse. The answer is only good because of the second half.
Gathered in parallel, and free
Every source is collected at once while you're still typing, so the extra context costs you nothing in wait time. Any source that fails is dropped in silence rather than blocking the answer.
Nothing leaves except the question
OCR and context gathering run entirely on your Mac. Only the assembled question and the region you circled make a network call, and follow-ups keep the whole thread, so "so is it cheaper than a boiler?" just works.
03<br>Output
Pixels in. Data out.
Every selection gets classified first, with a confidence and a reason, so you can see it understood a table as a table. When it finds a table or a chart, Make editable pulls it out of the pixels into a live editor you can type into and export.
What you circled
RegionQ3Q4Δ
Americas$1.24M$1.61M+29%<br>EMEA$880K$1.02M+16%<br>APAC$610K$79?K+3_%
Numbers · 604×188 · 9:41 AM
Detected: Table · 98% · four aligned columns with a header row
What changed between the quarters?
Americas grew fastest at +29%, and every region was up on Q3.
Make editable
Ask anything about this selection...
12 of 15 remaining
After Make editable
Table · 3 rows · 4 columns
Copy CSV<br>Export CSV
RegionQ3Q4Δ
Americas$1.24M$1.61M+29%<br>EMEA$880K$1.02M+16%<br>APAC$610K<br>unreadable<br>unreadable
+ Add rowEdit any cell. Values redraw as you type.
It never invents data. A cell the extractor couldn't read comes back empty and renders as an amber placeholder, never as a plausible guess. Chart values it had to infer from bar height, rather than read off a label, are marked approximate. Anywhere you see this color, in the app or on this page, it means the same thing: ScreenForm isn't sure, and it's saying so.
04<br>Artifacts
It doesn't stop at an answer.
Some questions want a sentence back. Others want a document, a working file, or something you can actually watch run. ScreenForm builds all three, from the same circle, and hands each one to the app that should own it.
"research this"
Heat pumps.pages
Heat Pumps in Residential Retrofit
Abstract
1 · Coefficient of performance
Sources
1. iea.org 2. nrel.gov 3. energy.gov
Research, into Pages
It searches your topic and several around it, writes a structured paper with an abstract, sections and cited sources, then opens it in Apple Pages ready to edit.
"write me the script"
cop_curve.py ~/ScreenForm/
1import numpy as np
3# COP falls as the gap widens
4def cop(t_in, t_out):
5 dt = t_in - t_out
6 return 6.2 - 0.09 * dt
7np.savetxt("cop.csv", curve)
Files, into your editor
Ask for a script, a component, or an SVG. It writes the file into ~/ScreenForm and opens it in VS Code, Cursor, Sublime, or whatever you use by default.
"animate how this works"
heat-pump-cycle.html<br>Open in new window
Previews you can run
Ask it to visualize or animate what you circled and it builds a working HTML and SVG preview. One button opens it full size in its own window, or in your browser.
It writes files. It never runs them. Everything it generates lands in one folder you can see, ~/ScreenForm , and opens in your own editor. Nothing is executed for you, so you always read the code before it does anything.
05<br>Capabilities
What it does once it has read.
Not a chat window you paste into. An action layer that sits over whatever you're already doing.
Answers stream in place
Responses render token by token in a panel beside your selection, formatted the way you would write them: headings,...