unitedrhino/docling is a pure Go implementation that follows the document model of the official Python Docling (Docling Core 1.10.0 protocol). It converts PDFs and Office documents, images, CAD drawings, and 3D models into a single DoclingDocument structure, forming the document ingestion layer used for RAG slicing and vectorization. The commit of October 14, 2026, 1a6eeb6, is a change that touches no code files and modifies only the two documents, README.md and README_EN.md.
The commit message is 'docs: README 지원 형식 목록 보완'. Two new tables were added to the front of the README. The input table lists PDF, .doc/.docx, .pptx, .xls (Excel 97–2003 BIFF8)/.xlsx/.csv, .epub, .chm, HTML, Markdown, AsciiDoc, EML, DWG/DXF, GLB 2.0, PNG/JPEG/BMP/WEBP/SVG, plain text, and DoclingDocument JSON re-reading, together with the supported scope for each format, and the output table pairs Markdown, HTML, DoclingDocument JSON, content_list JSON, and Outline with their respective API and CLI names.
The previous README listed the formats on a single line and the comparison table indicated only whether support existed, so limitations such as whether legacy .doc restores table boundaries and whether .xls computes formulas did not become apparent. The new input table states for each format both what it does and what it does not do, a GLB 2.0 row was added to the comparison table, and in the description of the differences from Python Docling, DOC, XLS, and CHM were moved from planned items to supported items.
That said, this commit documents what was already working rather than adding new functionality. The README makes clear that legacy .ppt is unsupported, legacy .doc extracts only paragraph and cell text, .xls reads only cached values and does not compute formulas, and GLB 2.0 extracts scenes, nodes, meshes, and materials but does not render geometry. OCR is not built in either, so scanned pages require OCRHook configuration, and the performance figures given in the README (Office documents median 4.4ms, papers 0.31s) are based on different measurement criteria, so the README itself emphasizes that they cannot be directly compared. Therefore this commit is a list to read when checking the supported scope, while whether to adopt it comes down to the remaining gaps: OCR and complex layout handling.