Pre-computed examples — demo NEVER fails. Three click-to-try buttons.

Client-side inference — browser calls HF directly, no server dependency

Back to InferenceClient — HF container DNS blocks direct HTTP

Fix build: match sdk_version 5.0.0, drop gradio from requirements

Fix API: api_name='run' + /gradio_api/call/run SSE pattern

Direct HTTP to HF API with token from env, no InferenceClient

Add HF_TOKEN to InferenceClient — Pro auth for priority inference

Fix lines: visible on white, gold stroke, full opacity on agree/disagree

Full light mode: white bg, dark text, gold accents, max readability

Node text: white, full opacity, text-shadow, bigger

Nodes visible: labels gold+bold, text white, opacity cranked

Lighter bg, brighter text, Triangulate button

Fix demo: multi-pattern Gradio API, no mode toggle, general only

Lighter: bg #1a1a22, white text, brighter nodes

Fix: brighter text, fix Gradio 5.x backend call via DOM

Full rewrite: lighter, show the process, agreement scores on lines, step-by-step narration

Fix layout: shrink triangle, kill centering, input box visible without scrolling

Fix build: sync sdk_version to 5.6.0, drop pyaudioop, keep Python 3.11

Fix Python 3.13 audioop error — add pyaudioop compat package