A Preview extraction pipeline turns DOCX, PPTX, and XLSX into Markdown for LLM and RAG pipelines. HTML input is not part of the pdfluent.extract_office call. Here is what is measured per format.
PDFluent can export a PDF to DOCX, XLSX, and PPTX (see the feature list). Converting the other direction — reading DOCX, PPTX, or XLSX and producing Markdown — is a Preview capability in PDFluent SDK 1.1.0. In Python, pdfluent.extract_office(path_or_bytes, format="markdown") takes .docx, .pptx, and .xlsx. HTML input is not part of that call. Install commands below carry no version. Structured extraction is Preview: output quality is still improving and the API may change in a minor release.
# Rust
cargo add pdfluent@1.1.0
# Python
pip install pdfluent==1.1.0
# Node.js
npm install @pdfluent/nodeA Preview extraction pipeline turns DOCX, PPTX, and XLSX input into Markdown, so you can feed Office documents into an LLM pipeline the same way you'd feed a PDF. In Python, pdfluent.extract_office(path_or_bytes, format="markdown") takes .docx, .pptx, and .xlsx. HTML input is not part of this call. Structured extraction is Preview: output quality is still improving and the API may change in a minor release.
import pdfluent
markdown = pdfluent.extract_office("deck.pptx")
docling_json = pdfluent.extract_office("sheet.xlsx", format="docling")Headings, paragraphs, lists, and tables in a .docx file map to their Markdown equivalents.
Slide text and titles map to Markdown; table structure and cell-level TEDS accuracy are not covered for PPTX (see the accuracy note below) — PPTX is evaluated on text, words, headings, heading levels, lists, and reading order, not tables.
Sheets map to Markdown tables. Word-level accuracy is not evaluated for XLSX (see the accuracy note below) — XLSX is evaluated on headings, tables, and table structure.
PDFluent's own claims register (CLAIMS.md) tracks exactly what was measured for each format, against what baseline, and what was explicitly excluded. Each claim below is scoped to specific metrics — do not read a claim about one format or metric as applying to another.
DOCX extraction is at least as good as Docling 2.132.0 on text, words, headings, heading levels, list levels, tables, TEDS, and coverage (93 public DOCX files from 23 websites; 95% confidence interval across websites). [CLAIMS.md A31, as of 2026-10-04]
PPTX extraction is at least as good as Docling 2.132.0 on text, words, headings, heading levels, lists, and reading order (68 PPTX files, 20 websites); tables and TEDS are explicitly not covered by this claim. [CLAIMS.md A32, as of 2026-10-04]
XLSX extraction is at least as good as Docling 2.132.0 on headings, tables, and TEDS (93 XLSX files, 25 websites); word-level accuracy is explicitly not covered by this claim. [CLAIMS.md A33, as of 2026-10-04]
For HTML, lists and tables are at least as good as Docling 2.132.0 (50 pages, 26 websites; the table lower bound is exactly 0.000 — i.e. not every table beats the baseline). Text and coverage are explicitly not covered by this claim, because the ground truth likely contains boilerplate that has not yet been checked. [CLAIMS.md A34, as of 2026-10-04]