Matteo Bortoletto

How we index images for RAG

Kapa converts documentation images into text captions at indexing time rather than sending pixels to multimodal models on every query. The approach improves answer quality while adding only 1–6% to per-query cost.

Jun 2, 20262 min read