https://arxiv.org/abs/2606.28344. The end of web parsing. The beginning of scalable pixel-native search. link: https://pixelrag.ai/
source:
type: kiwix
zim_path: wikipedia-simple # alias → auto-downloads Simple English Wikipedia (~1GB)
# Or use a file path: zim_path: ~/pixelrag-data/zim/wikipedia_en_simple.zim
# Or a URL: zim_path: https://download.kiwix.org/zim/wikipedia/...
ingest:
backend: cdp
quality: 85
embed:
model: Qwen/Qwen3-VL-Embedding-2B
device: cpu
output: demos/e2e/output