/root/warhammer-vision/
├── server.py          # FastAPI backend — sends images to LiteLLM vision model
├── requirements.txt   # Python deps
└── static/
    └── index.html     # Frontend — drag-drop, webcam, clipboard paste

Setup

cd /root/warhammer-vision
pip install -r requirements.txt

Run

# Default: uses gemma4-it:e4b on port 5173
python3 server.py

# Or customize:
PORT=5180 VISION_MODEL=ollama-lan python3 server.py

Env vars:

LITELLM_URL — default http://localhost:4000/v1/chat/completions
VISION_MODEL — default gemma4-it:e4b
PORT — default 5173

Status (2025-05-21)

Backend is built and tested — the server starts, serves the frontend, and routes images correctly through LiteLLM.

Vision model blocker — none of the vision-capable models work right now:

gemma4-it:e4b (FastFlowLM at 100.83.8.74:52654) — connection refused (service down)
ollama-lan (Qwen 3.6 on 100.83.8.74:11434) — times out on image requests (model too slow for 35B vision)

The app is ready. Once a vision model is working through LiteLLM, it'll just work.