fix: think parser starts in think mode; silent-reconstruction UI

Deploy: round-trip translator + decoder integration

Add starter prompt bubbles

Fallback thinking trace

cleanup: remove debug logging

Collapsible thinking traces + streaming

SHIP: fine-tuned emoji model live!

Adopt LiquidAI ThinkStreamParser + green palette

Fix parser for base model output

Switch to pipeline API + verbose errors

Revert to base model for working demo

Try fine-tuned ONNX model

Deploy Pantheon UI - emoji-only conversational AI

initial commit