cheap-scan1

To reduce API token consumption for academic PDFs, extract text locally using pymupdf and apply progressive triage. This filters content via abstract, introduction, and conclusion sections before selectively invoking vision models, minimizing expensive API calls.

Progressive triage filters PDF content sequentially through abstract, introduction, and conclusion sections. This funnel approach assesses relevance early, reducing token usage by only processing selective sections and routing specific patterns like tables to vision models when needed.

Yes, pymupdf is required for local zero-token text extraction and per-page quality metrics to detect scanned PDFs. It triggers fallbacks and optional page rendering for vision model analysis of tables, figures, and equations.

Safe mode performs exhaustive reads with per-section relevance tags for comprehensive analysis. Progressive mode uses a filtering funnel through abstract, introduction, and conclusion sections to minimize token usage during batch triage of working papers.

Yes, pattern-driven scanner routing automatically detects tables, figures, and equations within extracted PDF text. It optionally renders pages for parallel visual and equation analysis, producing consolidated markdown notes for downstream workflows.

Local text extraction cannot process scanned PDFs without triggering optional page rendering for vision models. Complex layouts may require selective visual extraction, and exhaustive safe mode reads increase token consumption compared to progressive triage filtering.