--- title: "liteparse" type: "AI Tool" url: "https://aidemos.com/tools/liteparse" description: "We tested LiteParse on native-digital reports: it preserved prose, headings, reading order, and TOC, but tables, charts, and scans degraded." category: "developer-tools" website: "https://github.com/run-llama/liteparse" published: "2026-07-21T07:41:11.650019+00:00" updated: "2026-07-21T15:24:53.247858+00:00" --- # liteparse Open-source PDF-to-markdown parsing that works well on native-digital reports, but degrades on scans, tables, and charts. ## TL;DR Verdict **Strong on clean reports, weak on scans and complex visuals** **Where it wins:** - You need a self-hosted, open-source PDF-to-markdown parser for native-digital reports. - You care most about preserving prose, headings, reading order, and table-of-contents structure on clean PDFs. - You can tolerate image placeholders and some cleanup in tables. **Main limitation:** Your PDFs are mostly scanned or OCR-heavy. **Pricing:** Free $0 `Native PDFs` · `Scanned OCR degrades` · `Tables fragile` · `TOC preserved` **Website:** [Visit liteparse](https://github.com/run-llama/liteparse) > **Strong on clean reports, weak on scans and complex visuals** > > LiteParse is a strong self-hosted parser for native-digital reports: it keeps prose, headings, reading order, and table-of-contents structure in good shape, and it at least preserves image placeholders instead of silently dropping assets on clean PDFs. The tradeoff is sharp: table structure is brittle, charts and figures are not faithfully reconstructed, and scanned layouts push OCR and image retention into clearly degraded territory. ## Demo Recording [Video: liteparse demo recording](https://d3epheqghktydj.cloudfront.net/liteparse-liteparse-input2-financialpdf-demo-3d371d0de95f.mp4) *Video — General walkthrough demo from the financial PDF test.* ## Feature-by-Feature Breakdown ### Text Extraction **Verdict:** Strong on native-digital prose; scanned text becomes noisy on complex layouts. LiteParse extracts readable prose from native-digital report pages, including long paragraphs, headings, and disclaimer copy, and can also OCR scanned research-paper pages with reduced reliability. The same capability is exercised on both the native reports and the scanned paper snippets, where OCR introduces substitutions, word splitting, and clipped lines. **Input:** > **Image** **Output:** > **Image** **Input:** **Output:** **Input:** **Output:** **Input:** **Output:** **Input:** > **Image** **Output:** > **Image** **Bottom line:** Clean on native PDFs; OCR degrades on complex scans, especially when layout and typography get harder. ### Table Extraction and Reconstruction **Verdict:** Usable on some clean tables, but headers, merged rows, and scan-heavy tables are fragile. LiteParse can turn some financial tables into usable markdown, but denser or scanned tables lose header structure, split multi-line labels, and drift out of row/column alignment. The evidence comes from the tested financial tables and scan-based table cases. **Input:** **Output:** **Input:** **Output:** **Input:** **Output:** **Input:** **Output:** **Input:** > **Image** **Output:** > **Text** **Input:** > **Image** **Output:** > **Text** **Input:** **Output:** **Bottom line:** Some cleaner tables are usable, but structural fidelity is inconsistent and scan-based tables are not dependable. ### Document Hierarchy Preservation **Verdict:** Strong on native report headings and TOCs; scanned titles and section headers fragment or flatten. LiteParse preserves markdown hierarchy on native-digital reports, keeping section headers and table-of-contents structure intact. On scanned paper, those same headings can collapse into plain text and lose their structural roles. **Input:** > **Image** **Output:** > **Image** **Input:** **Output:** **Input:** Source image > **File** — Source image **Output:** Parsed output > **Image** — Parsed output **Input:** Source image > **File** — Source image **Output:** Parsed output > **Image** — Parsed output **Input:** **Output:** **Input:** > **Image** **Output:** > **Image** **Bottom line:** Reliable for headings and TOCs on clean PDFs, but scanned titles and section headers lose their structure. ### Reading Order Preservation **Verdict:** Correct on clean multi-column and ordered layouts; scan-heavy pages can fragment flow. LiteParse usually keeps logical sequence intact in the CEO letter’s multi-column layout and the report’s ordered table of contents. The scanned paper test shows only partial breakdown, mainly through a fragmented list item that interrupts flow. **Input:** **Output:** **Input:** **Output:** **Input:** **Output:** **Bottom line:** Good at preserving logical flow in native PDFs, with only partial breakdown on the scan test. ### Embedded Image Preservation **Verdict:** Native-page images are kept as placeholders, but scanned visuals can disappear entirely. On native PDFs, LiteParse keeps portraits, signatures, and logos as image placeholders or references. On scanned pages, the visuals can be dropped entirely, showing the same image-retention behavior under weaker source conditions. **Input:** > **Image** **Output:** > **Image** **Input:** > **Image** **Output:** > **Image** **Input:** **Output:** ``` The USDA logo is omitted entirely from the extracted output, with no placeholder or markdown image syntax generated. ``` **Input:** > **Image** **Output:** > **Text** **Bottom line:** Native images are technically retained, but scanned visuals are dropped outright. ### Chart and Figure Reconstruction **Verdict:** Charts and figures are not faithfully reconstructed; visual meaning is lost. LiteParse attempts to preserve or reconstruct charts and figures from the tested report pages, but the observed outputs often fall back to placeholders or garbled text. The behavior is shown on both the Target report chart and the scanned research-paper figure. **Input:** Source image > **File** — Source image **Output:** Parsed output > **Image** — Parsed output **Input:** Source image > **File** — Source image **Output:** Parsed output > **Image** — Parsed output **Bottom line:** Chart/figure reconstruction is one of LiteParse’s weakest areas. ### Caption/Figure Association **Verdict:** Captions stay attached when the source layout is intact, but become orphaned when visuals vanish on scans. LiteParse keeps captions and nearby explanatory text attached to the right table or figure on native PDFs, and also keeps caption-like text linked to associated visuals when the page structure is intact. When scanned visuals disappear, those associations break and the text becomes detached. **Input:** > **Image** **Output:** > **Image** **Input:** Source image > **File** — Source image **Output:** Parsed output > **Image** — Parsed output **Bottom line:** Works on clean pages, but scan-heavy documents leave captions detached from their visuals. ### Long-Document Robustness **Verdict:** Quality stays consistent across document length; source type is the real separator. LiteParse maintains a consistent baseline across the 84-page hybrid earnings report, the 18-page financial report, and the 12-page scanned paper. The test results suggest document length itself is not the limiting factor; source quality is. **Input:** **Output:** **Input:** **Output:** **Input:** **Output:** **Bottom line:** Document length itself was not a problem; the source type determined whether output quality stayed strong or fell apart. ## Open source | Plan | Price | Notes | | --- | --- | --- | | Free | $0 | Open source | ## Is It Right For You? **Use it if** - You need a self-hosted, open-source PDF-to-markdown parser for native-digital reports. - You care most about preserving prose, headings, reading order, and table-of-contents structure on clean PDFs. - You can tolerate image placeholders and some cleanup in tables. **Skip it if** - Your PDFs are mostly scanned or OCR-heavy. - You need faithful chart and figure reconstruction. - You need dependable complex table reconstruction with merged headers and split rows. ## Classification - **Category:** developer-tools - **Subcategory:** documentation-tools - **Type:** text ## Frequently Asked Questions **Q: Does LiteParse work better on native PDFs or scans?** It worked much better on the two native-digital reports. Text, headings, reading order, and table-of-contents structure stayed strong there, while the scanned paper showed OCR garbling, broken tables, omitted images, and fragmented headings. **Q: How well does LiteParse handle tables?** Mixed. Some cleaner tables were usable, but headers flattened and rows split on the hybrid earnings and financial report, and the scanned paper tables were severely degraded or unusable. **Q: Does LiteParse preserve headings and table of contents structure?** Yes on native-digital reports. Section headings and TOC ordering were preserved with real markdown hierarchy, but scanned titles and section headers fragmented or flattened. **Q: What happens to images and figures?** On native PDFs, LiteParse keeps images and signatures as markdown image placeholders, but they are not actually viewable. On the scanned paper, the USDA logo and other visuals were dropped entirely. **Q: Can LiteParse keep reading order in multi-column PDFs?** Yes in the CEO letter test, where it read the columns in the correct sequence. The scanned paper was only partially reliable, with at least one fragmented list item disrupting flow. **Q: Is LiteParse free and open source?** The report identifies LiteParse as open source and lists it as free at $0. ## Similar Tools AI tools similar to liteparse: - [PyMuPDF4LLM](https://aidemos.com/tools/pymupdf4llm) — Open-source PDF-to-markdown for clean native PDFs, but unreliable on scans, dense tables, and images. - [doc2mark](https://aidemos.com/tools/doc2mark) — Open-source PDF-to-markdown that preserves native text and headings, but still struggles with tables, charts, images, and scans.