Currently under development on Mac OS X (note: transparency now done manually, so no need for OpenGL...)

Based on the Dasher codebase, so uses all alphabets (/colour schemes), with e.g. game mode removed, as well as most input filters (Click mode zooms *2 on any point, press key to zoom back out; normal filter zooms in if in middle 1/4, else zooms+translates, but would be better to behave just like click mode with separate backoff key).

Transparency: make node "height" (in Z axis, in/out of screen) proportional to probability; then fill with uniform-density "fog" of node colour.

Output: currently the deepest node covering 1/2 the screen pixels is selected, but this needs more work.

Initial view, with lines drawn between consecutive letters; observe capitals in yellow...

Alternative view, no lines between letters, but "prepending" all unentered text (this screen shot shows having just written "United ", about to proceed to "States")...