Local-first fork of T3 Code with GPU server registry and dark theme

Mia@MiaAI_lab

I made some changes to T3 Code to make it local-first.

A new page in Settings > Inference, allows you to directly add your local models, and then attach it to Pi (or others), so it's not depended on what models are configured in Pi on your host.

I've also added an actual Dark theme, and made the previous dark theme as OLED. "Dark" theme was too black for me.

I'll continue modifying it and adding more features.

Full list of changes:
- Local models are the default, cloud is the backup
- Register your own GPU servers, live health & cache
- Link Pi / OpenCode / Codex to a server in one click
- A gateway queues requests, so threads never swamp the GPU
- Automatic fallback when a server goes down
- Titles, commits & PR text written by your local model
- Per-server stats: tokens, cache hits, TTFT, tok/s
- Local runs show "Local", never a fake $ cost
- Lean tool set: ~40% fewer prompt tokens for local models
- Two-tone Dark theme + true-black OLED

Get my fork here: