Recursive AI
LibraryC#Available

onedocument

Word, Excel and PowerPoint documents in .NET, in C#. Read, write, edit and render DOCX, XLSX and PPTX, the legacy binary formats, ODF, RTF, HTML and Markdown, with no Office and no native code.

Open source, MIT or Apache-2.0, made by Recursive AI. Version 1.0.0 is on NuGet. Formerly OneOffice; not affiliated with Microsoft.

dotnet add package RecursiveAI.OneDocument.All
Getting started Docs GitHub NuGet

A quick look

Open a document or a workbook by its content, read what’s in it, and convert it to a tagged PDF in one line.

The output shown is from a run on 1.0.0 from NuGet, on a report and a budget the program made itself. Getting started has the whole program.

Getting started
Program.csC#
using (Document doc = Document.Open("report.docx"))
    foreach (Paragraph p in doc.Paragraphs
             .Where(p => p.Style == "Heading 1"))
        Console.WriteLine(p.Text);
// Findings
// Next steps
 
Office.Convert("report.docx", "report.pdf");
// a one-page tagged PDF
 
using Workbook wb = Workbook.Open("budget.xlsx");
Cell total = wb.Sheets["Budget"]["D2"];
Console.WriteLine($"{total.FormattedText} = {total.Formula}");
// $591.50 = SUM(B2:B4)
What it does

The formats

Read and write DOCX, XLSX and PPTX (with macro-enabled, template and Flat OPC variants), XLSB, XLS (BIFF8), RTF, ODT, ODS, ODP, CSV, HTML and Markdown. Read only: DOC, PPT, older XLS (BIFF2 to BIFF5) and Word and Excel 2003 XML. Encrypted files open with their password.

Every format

Edits that leave the rest alone

In the README’s words: “Parts you do not touch are written back byte-identical (raw zip entries, never recompressed); edits keep unknown elements, attributes, prefixes, mc:AlternateContent and extLst in order.” Damage is repaired and reported as a located defect.

Contracts

PDF and page images

Documents, workbooks and presentations to tagged PDF, with PDF/UA-1 or PDF/A-2a profiles, and to PNG, JPEG, TIFF, WebP and SVG pages. Layout follows Word, Excel and PowerPoint’s rules and is measured against Office’s own PDF export.

Layout and PDF

Calculation

A calculation engine with Excel’s semantics: dynamic arrays and spill, LET and LAMBDA, structured and 3D references, iteration, data tables, Goal Seek and pivot refresh, with Excel’s number formats. Formula text “is en-US and invariant only; localized function names are not supported.”

Recalculation

Templates and operations

One template engine across Word, Excel and PowerPoint files, mail merge and content-control binding. Merge, append, split, find and replace across runs and cells, tracked changes, protection, and rows of plain objects in and out of a workbook.

Templates

Inspect and sanitize

Find and remove hidden data and personal information: comments, revisions, properties, hidden sheets, external links and more, in Document Inspector’s categories. Some categories are reported but not removed. Not independently audited.

Inspect and sanitize

Redaction

Removes matched text and regions from the file, not only from view, writes a new file and re-reads every entry of it to verify. Matching is exact on the characters, so variants must be listed. Not independently audited.

Redaction

VBA as data

Read, list, extract and remove VBA projects. In the docs’ words, VBA “is never compiled, interpreted or run, the p-code is never decoded, and no password is cracked.” Removal is not independently audited.

VBA projects

Accessibility and platforms

An accessibility check against Office’s published rules; agreement with Office’s own checker is pending, because it has no automation interface. NativeAOT, trimming, browser WebAssembly, iOS and Android, except encrypted files in the browser (below).

Platforms
How it measures

What was checked for 1.0

The project’s own checks (not an external audit), each with the note that holds it, in the overview.

  • Untouched round trip. Over 4,900 public OOXML and XLSB files, “111,239 of 111,239 parts byte-identical at the package layer; 6 files refused with a typed defect.”
  • Office opens the output. On desktop Office 16.0.20527, “718 generated files open clean” in Word, Excel and PowerPoint, with no repair prompt.
  • Calculation. Recalculation reproduces the cached values of 99.52% of the testable formulas in 354 Common Crawl workbooks and 93.02% in 702 bug-tracker workbooks; against Excel’s own results, function rows agree 93.3%, financial 84.5% and engineering 88.0%, with every disagreement listed.
  • Layout against Office’s PDF export. Page count agrees on 196 of 196 documents and workbooks and 24 of 24 decks. Mean text-line agreement: business documents 0.951, long documents 0.966, edge features (floats, drawings, diagrams, math) 0.834, worksheets 0.934, presentations 0.748.
  • Speed, as a ratio. OneDocument’s median time divided by the fastest like-for-like managed library’s, over 49 rows, single thread, so below 1 means OneDocument took less time: geometric mean 0.30 on x64 and 0.32 on arm64. The slowest rows are 1.08 on x64 (writing a 10,000-row XLS, against NPOI) and 1.07 on arm64 (a streaming read of 10,000 rows, against Sylvan.Data.Excel).
  • Redaction and sanitizing reviews. Both are the project’s own reviews (not an external audit). The redaction review: “After the fixes in this review there is no tested input from which a matched string can be read back and which the report calls verified, except the four where the plan’s literal does not match the text at all.” The sanitizer review: “Every removal the report claims holds in the 61 passing cases.” Three gaps remain, listed in the review.
The 1.0 gates
Limits

Known gaps at 1.0.0

  • Encrypted files in the browser. “WebAssembly has no Aes or MD5, so opening Agile or Standard encrypted OOXML and RC4-MD5 files there raises OfficeUnsupportedException. RC4 CryptoAPI and XOR files read normally.”
  • Accessibility Checker. “Pending: Office has no automation for it, so our results are checked against the published rule set only.”
  • Signatures and encryption on write. RecursiveAI.OneDocument.Cryptography “has no content yet and is not in RecursiveAI.OneDocument.All; it follows in a 1.x release.” Until then, encrypted files are written only for XLS.
  • Formula text “is en-US and invariant only; localized function names are not supported.”
  • Document Inspector. Office can be asked about only some categories, so for comments, revisions, properties, embedded documents, macros and links, agreement “was not measured; our own re-inspection finds nothing left in sanitised output.”
  • Read-only formats. “DOC and PPT are read-only.” Excel doesn’t open the ODS files OneDocument writes, though they validate and LibreOffice opens them.
  • PDF/A-2a. “19 of 20 DOCX and 15 of 16 PPTX pass; the failures are one rule” (a private-use character without ActualText).
Every gap, in the overview

Status and packages

Go further

GitHub NuGet Getting started All docs Coming from another library Checks against Office Changelog

OneOffice is now OneDocument. OneDocument is not affiliated with Microsoft or with any other product or company named OneDocument; Word, Excel and PowerPoint name the file formats.