Save ideas, not pictures

Every diagram tool eventually faces the same temptation: when a user pastes a screenshot, just trace it. Match the boxes, match the arrows, ship it. We tried that first. The results looked impressive for about four seconds — right up until you tried to edit them.

A traced diagram is a photocopy. It inherits every accident of the original: the box that was too small because someone ran out of whiteboard, the arrow that bends three times for no reason, the inconsistent gaps. Worse, the copy has no idea what it depicts. Move one box and nothing follows.

Meaning first, pixels never

When you paste into sket, the model is explicitly forbidden from emitting coordinates. It produces an intermediate representation — nodes, edges, nesting, emphasis — and nothing else. Press play below; this diagram will walk you through the pipeline itself.

The layout engine that turns IR into shapes is deterministic and entirely boring, in the best way. Layered flows read left to right. Containers nest. Primary nodes grow. Databases get a database glyph. The same intent always produces the same drawing — there is no model in the drawing loop, so there is no slot-machine energy in the output.

Why this matters for you

• Everything stays editable. The output is ordinary sket shapes. Drag them, retheme them, delete half and keep going.
• Your style applies. The IR doesn't know what theme you use, so the drawing lands in your feel, on your grid — not a pastiche of the screenshot's.
• It round-trips. Copy any sket diagram and the clipboard carries plain JSON shapes with a version header. Paste it anywhere, including back.

The screenshot you pasted was the last time that idea had to live as pixels.

Try it: open the canvas, paste any architecture screenshot, and watch what survives — the meaning, not the mess.