Skip to main content
The demo we reach for when someone asks “what can Glasswarp do that a VNC or a throwaway sandbox can’t?” An agent turns a source portrait into a paint-by-numbers plan and paints it in MS Paint on a real Windows PC through the API.
Platform vs agent: Glasswarp provides eyes (screenshot, canvas/ribbon UIA targets) and hands (drag, click, native color-dialog input). Turning an image into a tile plan is the agent (customer code).

​
Why it’s a differentiator

1

Exact color through UI structure

Each tile color is set via Paint’s native Edit Colors dialog using host UIA targets (the R/G/B fields). A pixel-only VNC framebuffer has no UI tree to target, so it can’t reliably dial in precise RGB.
2

High-volume precise native input

Hundreds of native drags (the tile grid) and native bucket-fills land 1:1 on the real Win32 canvas at native resolution. Soft framebuffers smear and misplace this at scale.
3

Native-resolution verification

The result is captured as a fresh, native-res GPU screenshot — not a re-encoded remote frame.

​
Prerequisites

1

An online, API-enabled rig

See Install the host.
2

A host with drag support

The host must be rebuilt with mouse_down / mouse_up.
3

SDK with grounding extra

​
Run

A public-domain Mona Lisa ships at sdk/python/examples/assets/mona_lisa.jpg.

​
Tuning

Defaults produce a ~28×42 ≈ 1176-tile, 12-color mosaic, letterboxed so tiles stay square. When finished, the agent uses Paint File → Save As to write a JPEG onto the Windows rig Desktop: %USERPROFILE%\Desktop\Mona Lisa via Glasswarp API.jpg Watch the Save dialog in Console → Sessions → Eye.

Paint the wordmark

The themed GLASSWARP headline visual.

Send input

Drag, click, and native color selection.