Use codex to fix some errors?

[sheikhti1205]

I was doing a project for my university, then found this awesome app. But had some flaws. And for personal reasons I really wanted to use android to handle the server part for my project. But I encountered many issues on this app. So, I felt like fixing those using codex for my personal use. But now I was thinking if the devs are going to be interested in viewing the codex edited contents. If so I'll generate a note on what changes it made and leave a personal opininon on what worked and what didn't. On second thought, I'll list them all here:
ToolNeuron change list

1. Server model selection fixed

• Added stricter server-side model handling.
• Server no longer silently auto-picks the first available model.
• Chat requests now require a valid chat model, or use the configured chat default.
• Wrong model IDs now return an error instead of falling back.
• Added Android Server screen chat model picker.
• The server chat picker updates the actual server default model config.
• Bundled Web UI Settings now has a real chat model dropdown from /v1/models.
• Web UI chat model dropdown only shows Chat/VLM models, not embeddings/upscalers.
• Fixed the bug where an embedding model like nomic-embed could be selected for chat.
• Server still supports role defaults for embeddings, voice, image gen, upscaling, etc. through Settings > Server model roles.

2. Manual model categories added

• Installed models can now be manually assigned to categories.
• Supported categories:
  Chat
  Embedding
  Image generation
  Image upscaler
  TTS
  STT
• The app uses these assigned categories throughout Store, Model Manager, Server, and Image Task screens.
• This removes reliance on fragile auto-detection for special models like upscalers.

3. Store and model management fixes

• Installed tab now still shows Import local model even when no models are installed.
• Added better image generation and image upscaler handling.
• Fixed store filtering/type inference issues.
• Added more useful default targeted repos for use cases like embeddings and coder/chat models.
• Local model imports now copy files into app storage instead of depending on fragile content URIs.

4. Image task improvements

• Added progress/metrics card for Generate, Inpaint, and Upscale.
• Generate/Inpaint show real native step progress.
• Metrics include elapsed time, ETA, percent complete, current step, step time, and output size.
• Upscale shows elapsed time, estimated ETA/progress, and final native processing time when complete.
• Added output behavior options:
  Keep result in session
  Replace input image
  Auto-save to Android Photos at Pictures/ToolNeuron
  Save elsewhere using Android save dialog
• Result cards now include both Save Photos and Save as options.

5. Web UI redesign

• Redesigned bundled browser UI to look closer to ToolNeuron app.
• Changed from generic green web UI to dark blue app-style panels.
• Added more rounded cards, app-like sidebar, larger controls, blue accent styling, and a closer mobile-app visual language.
• Settings modal now matches the updated server model selection flow.

Known limitation

• Native upscaler progress is still estimated while running because the current native upscaler only reports Processing and Complete, not real step-by-step progress. Final processing time is real.

[ZHINFINITY]

Incredible, wonderful

[Siddhesh2377]

Hey @sheikhti1205 I would Love this Changes Please Open a PR

[sheikhti1205]

Ok, I will