fix: Gradio 5.50.0 (supports Python 3.13, no pydub/audioop dep)

README.md

@@ -4,7 +4,7 @@ emoji: ✨
 colorFrom: yellow
 colorTo: gray
 sdk: gradio
-sdk_version: 5.0.2
+sdk_version: 5.50.0
 app_file: app.py
 pinned: false
 tags:
@@ -22,7 +22,7 @@ tags:
 
 **A daily ritual where your future self sends you transmissions.**
 
-Check in with one word. Receive a personalized voice transmission from across time. Make a tiny choice that reshapes who gets to speak tomorrow.
+A private future-radio, not a dashboard. Check in with one word, receive a voice transmission from across time, make a tiny choice that reshapes who gets to speak tomorrow. The whole ritual lives inside a single signal chamber — no card-grid, no SaaS, no scroll.
 
 All inference runs on-device via three small models — no cloud dependencies, no API bills, no data uploaded.
 
@@ -54,13 +54,13 @@ Each model is well under 32B params. Total: ~3.1B across all three models — qu
 | **NVIDIA Nemotron** | Nemotron-Parse for structured insight extraction from user notes |
 | **Tiny Titan** | ~3.1B total across all models — genuinely tiny |
 | **Best Agent** | Multi-step agentic pipeline: check-in → extract → generate → choice → reaction → persist |
-| **Off Brand** | Custom Gradio CSS with dark amber theme, card-based layout, animated loading state |
+| **Off Brand** | Custom transmission-console interface (Fraunces + IBM Plex Mono), atmospheric signal styling with layered gradients, animated tuning-state waves, voice-orb constellation rail, and a horizontal signal-path progress bar — no card-grid dashboard |
 | **Best Demo** | Full demo video + social post (links below) |
 | **Bonus Quest Champion** | Targeting 6+ bonus/sponsor criteria simultaneously |
 
 ## Tech
 
-- **UI:** Gradio 5 with custom CSS theme (Off Brand)
+- **UI:** Gradio 5 with a custom transmission-console skin (Fraunces serif for the chamber voice, IBM Plex Mono for instrument labels), atmospheric gradients and scanlines, a horizontal voice-orb constellation rail, and a signal-path progress bar in place of step pills
 - **LLM:** MiniCPM 2.5 via 🤗 Transformers with torch.compile + SDPA attention
 - **Extraction:** Nemotron-Parse (NVIDIA) with keyword fallback when GPU is constrained
 - **TTS:** Kokoro 82M — generates WAV output for each transmission

app.py

@@ -1050,11 +1050,14 @@ def _render_transmission(state: AppState) -> str:
     title = _chamber_title(f'<em>{t.title}</em>')
     body_html = _chamber_body(t.text)
     actions = _audio_module(label, state.today_audio) if state.today_audio else ""
+    # The chamber holds the audio module + tonight's move. Tomorrow's cliffhanger
+    # is shown as a second anchor line below the chamber so the page has a clear
+    # rhythm: text → audio + tonight → tomorrow → constellation → memory log.
     tonight = _chamber_callout("tonight's move", t.action_prompt)
     tomorrow = _chamber_callout("tomorrow", t.cliffhanger)
     out = _identity_strip(state)
     out += _signal_chamber(eyebrow, title, body_html, actions_html=actions, callout_html=tonight)
-    out += _chamber_callout("tomorrow", t.cliffhanger)  # shown as a second anchor line below chamber
+    out += tomorrow
     out += _constellation_rail(state)
     out += _signal_path("choose")
     out += _memory_log(state, limit=5)
@@ -1095,6 +1098,7 @@ def _render_history(state: AppState) -> str:
         out += _signal_chamber(eyebrow, title, body)
     out += _constellation_rail(state)
     out += _memory_log(state, limit=20)
+    out += _signal_path("react" if state.choice_made else "check-in")
     out += _signal_footer()
     return out
 

requirements.txt

@@ -1,8 +1,6 @@
 # FutureSelves — Build Small HF Space
 # Core
-gradio==5.0.2
-# Python 3.13 compat: audioop removed in 3.13, needed by pydub/gradio 4.x
-# Gradio 5.x uses its own audio handling
+gradio==5.50.0
 torch>=2.2
 transformers>=4.40
 sentencepiece>=0.2