Give a companion character sub-second typed reflexes

01 · Role in the system

What Jev does here

Every touch - hair, forehead, cheek, shoulder, collar - fires a reflex pass in which Jev returns, in one call, her true feeling (Choice), how affection moves (Score), whether she disliked it (Noul) and the next topic (Choice). Face and a fixed one-liner render from that answer in 0.2 to 0.5 seconds; Claude then writes the full spoken line from the feeling, and Gemini TTS streams the voice about a second later, its tone colored by feeling, scene, intensity and familiarity. All four channels anchor to the same judgment, so face, one-liner, dialogue and voice never contradict even though they arrive at different speeds. Context shifts meaning - the same shoulder touch after she shows weakness reads as worry - and scenarios branch across five stages, with a keyless public demo rate-limited to 10 minutes and 30 moves per person.

02 · Control boundary

Where Jev sits

Jev as the sub-second reflex layer in front of generative models: typed judgments drive the immediate reaction, and the slower LLM dialogue and TTS voice are constrained to stay consistent with the judgment that already landed.

Code owns the loop, permissions, thresholds, validation, and side effects. Jev owns only the bounded judgments described above.

03 · Known limits

What this evidence does not prove

• Beta; a single scripted character, Japanese only, with affection mechanics designed by the author.
• The keyless public demo is rate-limited per person and per day.
• Reaction latency is author-reported; no independent timing data.

04 · Attribution

Public sources

This is a Community record: the project was published by a third-party community author.

• ozaki-taisuke ↗Community · github · public · checked 2026-09-27