Pure text glyphs. Mouth = viseme glyph, eyes blink + track gaze, thought bubble while thinking, transcript runs as a mouth-ribbon while speaking. Near-zero render cost — works anywhere text renders, even a terminal.

The transcript IS the face: the current word is the mouth — it stretches with the actual audio energy, pops on emphasis (!), drifts up as speech flows. Eyes live above the words. Needs a transcript lane (live agent or the word feed).

Continuously-redrawn ink character — the stroke "boils" like hand-drawn animation (re-jittered ~11×/s). Parametric mouth (width/openness from visemes + energy), brows carry emotion, head tilts with the idle sway. The sketch-host direction from the P18 pack.