Retry transient VoxCPM accelerator errors

Add VoxCPM reference audio cloning

Add VoxCPM ZeroGPU voice provider

Default portraits to Klein ZeroGPU

Add Klein ZeroGPU portrait option

Portraits: in-browser WebGPU via vendored Bonsai (FLUX.2-Klein 4B, on-device)

Portraits step 4: in-browser Janus-Pro-1B (WebGPU) portrait engine

Portraits step 3: persona panel Portrait section + per-hero caching

Portraits: set correct Content-Type for cloud images (NIM returns JPEG)

Portraits step 2: /portrait engine routing + imagen facade + Settings picker

Settings: hide the app URL printed in Gradio's Settings header

Settings: "Use recommended for this device" button (GPU/device-tier heuristic)

Settings presets: real padding (beat Gradio reset) + readable active hover

Settings: recommended quality presets + device readout; persist all choices

Voice: default prod provider to Kokoro (in-browser), not cloud Qwen3-TTS

Voice: provider-driven per-hero voices; settings is provider-only

Persona: simplify prompt to name/about/quote/voice; editable in Settings

Persona voice: re-DESIGN when the voice description changes (not clone)

Persona voice: play↔stop toggle, loading spinner, auto-stop on nav/new pick

Persona voice: cloud voice-clone, varied voice designs, animated class picker

Personas → heroes: shield icon, hero terminology, empty state, active highlight

Fix: red headings, voice-edit badge, and voices persisting across refresh

Persona voice: one pulsing play button, sidebar-style headings, live badge

Persona: red section headings, icon buttons, edit-aware voice + clone-on-play

Persona: editable fields (auto-save), quote after about, Create/Replay voice

Personas-as-agents: spoken quote + save to a local-first barracks roster

Voice: cloud Qwen3-TTS is the default engine in prod (was Kokoro)

Voice settings: add "Qwen3-TTS local" engine — localhost-only, default on localhost

app.py: PORT env (default 7860) so it runs locally next to a service on 7860

Add Qwen3-TTS local bridge: ?tts= override + tts_server.py

Move Voice settings into the Settings page; add a sidebar ⚙ Settings button

Qwen3-TTS: use DashScope international endpoint (our key is sk-ws- intl)

Add Qwen3-TTS Voice Design (DashScope) + persona voice descriptions

Cap persona + diary at 100 tokens; shorten prompts; stream the diary

Fix empty-JSON: keep the answer when a model leaves <think> unclosed

Model bar: show params once (avoid '0.6B · 0.6B' for byte-less engines)

Settings: model section to top; default WebLLM + Qwen3 0.6B; copyable debug

Put the model picker in Gradio's OWN settings page (not a custom tab)

Move engine/model settings to a Settings page; share one model across both pages

Cache status + delete for WebLLM and Transformers.js (like wllama)

Transformers.js: pre-flight WebGPU device, else WASM (fixes load + caching)

Fix stale JS (no-cache on /web) + Transformers.js WebGPU→WASM fallback

Fix Qwen3 on WebLLM: budget for thinking + capture reasoning_content

Fix Qwen3 parsing (/no_think) + WebLLM double-load guard & progress clarity

Mobile: collapse model/voice control bars behind tap-to-expand

Mobile: give persona/diary content real space on phones

Add in-browser TTS: read war diaries aloud (Kokoro / Kitten / Web Speech)

Fix model re-downloads (persistent storage) + WebLLM shader-f16 crash

Multi-engine benchmark: wllama vs Transformers.js vs WebLLM (engine selector)

Add Qwen3-0.6B; surface WebGPU backend; strip <think> from answers