Which free tools and methods recover tables from scanned PDFs, JavaScript-only filing viewers, and chart images? For each: what it misses, one example document type, and whether it is local and free, hosted and free within a limit (state the limit), or paid.
A curated shortlist of 12 tools and methods, consolidated from 49 evidence rows across independent reviews, benchmarks, vendor docs and papers, grouped into the three input classes. Current as of 2026-11-03; free tiers and pricing can change.
Run Docling or PaddleOCR locally — fully on-prem, air-gap compatible, no data leaves the machine.
Go to EDGAR/XBRL first — the API or the raw inline-XBRL URL; browser automation only as a fallback.
WebPlotDigitizer with axis calibration and manual point correction.
Model-assisted extraction (DePlot, ChartRecover) plus sampled manual verification.
| Tool | Input class | Access | Misses | Example document | Source |
|---|
Method: curated shortlist of 12 tools consolidated from 49 rows across 5 result sets (tool reviews, benchmarks, vendor release notes, papers); duplicate rows for the same tool merged into one card with its strongest-sourced caveat. Each card states how table recovery works, the key miss, one example document type, and exactly one access label; hosted commercial services with no stated free allowance are classed as paid. Accuracy figures come from the cited benchmarks, not our tests. Review current as of 2026-11-03; free tiers and pricing can change. Secondary capabilities and duplicate source variants cut for space.