OfficeMD

Fast Office document extraction for LLMs and agents. Converts DOCX, XLSX, CSV, PPTX, and PDF into clean markdown, structured JSON IR, and Docling output.

• Native Rust core - fast, no runtime dependencies
• Three output modes: markdown, structured JSON IR, Docling JSON
• CLI and SDK for Python, Node/Bun, and Rust
• Sheet, slide, and page selection
• Document property extraction

Quick Start

No install needed - run directly:

uvx officemd markdown report.docx
npx office-md markdown report.docx
bunx office-md markdown report.docx

Install

Homebrew (macOS / Linux)

brew tap thomaub/officemd [email protected]:ThomAub/officemd.git
brew install thomaub/officemd/officemd_cli

Shell / PowerShell (from GitHub Release)

Installers generated by cargo-dist:

# macOS / Linux
curl --proto '=https' --tlsv1.2 -LsSf https://github.com/ThomAub/officemd/releases/latest/download/officemd_cli-installer.sh | sh

# Windows (PowerShell)
powershell -ExecutionPolicy ByPass -c "irm https://github.com/ThomAub/officemd/releases/latest/download/officemd_cli-installer.ps1 | iex"

Python

uv tool install officemd

Or add as a dependency:

uv add officemd

Node / Bun

npm install office-md
# or
bun add office-md

Rust

cargo install officemd_cli

CLI

All three surfaces expose a CLI named officemd (Python, Rust) or office-md (Node/Bun).

officemd markdown report.docx
officemd markdown budget.xlsx --sheets "Summary,Q1"
officemd markdown deck.pptx --pages 1-3
officemd render report.docx
officemd diff old.docx new.docx

The Rust CLI has additional subcommands:

officemd stream report.docx                    # stream to stdout (supports stdin via -)
officemd convert report.docx --output out.md   # write to file
officemd inspect report.pdf --output-format json --pretty

Common options

Flag	Description
--format	Force document format (docx, xlsx, csv, pptx, pdf)
--pages	Select pages/slides/sheets by index (e.g. "1,3-5")
--sheets	Select sheets by name or index (e.g. "Sales,1-2")
--include-document-properties	Include document metadata in output
--markdown-style	Output style: compact (default) or human

SDK

Python

from pathlib import Path
from officemd import extract_ir_json, markdown_from_bytes, docling_from_bytes

content = Path("report.docx").read_bytes()

print(markdown_from_bytes(content, format="docx"))
print(extract_ir_json(content, format="docx"))
print(docling_from_bytes(content, format="docx"))

Python patch reports and batch patching

from pathlib import Path
import officemd

content = Path("report.docx").read_bytes()
patch = officemd.DocxPatch(
    scoped_replacements=[
        officemd.ScopedDocxReplace(
            officemd.DocxTextScope.ALL_TEXT,
            officemd.TextReplace("word", "term"),
        )
    ]
)
# ALL_TEXT includes document content plus free-text metadata/app/custom fields.

single = officemd.patch_docx_with_report(content, patch)
print(single.report.replacements_applied)

batch = officemd.patch_docx_batch_with_report([content, content], patch, workers=4)
for item in batch:
    print(item.report.parts_scanned, item.report.parts_modified, item.report.replacements_applied)

Rust batch patching with reports

use officemd_core::{
    patch_docx_batch_with_report, DocxPatch, DocxTextScope, ScopedDocxReplace, TextReplace,
};

let patch = DocxPatch {
    set_core_title: None,
    replace_body_title: None,
    scoped_replacements: vec![ScopedDocxReplace {
        scope: DocxTextScope::AllText,
        replace: TextReplace::all("word", "term"),
    }],
};
// AllText includes document content plus free-text metadata/app/custom fields.

let results = patch_docx_batch_with_report(vec![doc1_bytes, doc2_bytes], &patch, Some(4))?;
for item in results {
    println!(
        "parts_scanned={} parts_modified={} replacements_applied={}",
        item.report.parts_scanned,
        item.report.parts_modified,
        item.report.replacements_applied
    );
}
# Ok::<(), officemd_core::PatchError>(())

Patch scopes also support free-text metadata/comment fields:

• DOCX: MetadataCore, MetadataApp, MetadataCustom, MetadataAll
• PPTX: CommentAuthors, MetadataCore, MetadataApp, MetadataCustom, MetadataAll
• XLSX: Comments, CommentAuthors, MetadataCore, MetadataApp, MetadataCustom, MetadataAll

AllText now means all free-text fields: content + metadata/comment-author text.

Formatting-preserving replacement is available for OOXML content text:

use officemd_core::{DocxPatch, DocxTextScope, ScopedDocxReplace, TextReplace};

let patch = DocxPatch {
    set_core_title: None,
    replace_body_title: None,
    scoped_replacements: vec![ScopedDocxReplace {
        scope: DocxTextScope::Body,
        replace: TextReplace::all("Confidential", "")
            .with_preserve_formatting(true),
    }],
};

Semantics:

• replacement can span multiple runs
• the first matched run keeps the replacement text and therefore its formatting wins
• later consumed runs are left empty in v1
• metadata/comment-author fields still use simple text replacement

JavaScript

import { readFileSync } from "node:fs";
import { markdownFromBytes, extractIrJson, doclingFromBytes } from "office-md";

const content = readFileSync("report.docx");

console.log(markdownFromBytes(content, "docx"));
console.log(extractIrJson(content, "docx"));
console.log(doclingFromBytes(content, "docx"));

WebAssembly demo

There is also a browser demo for the WASM bindings in crates/officemd_wasm/README.md. It serves a small page at http://localhost:8080/crates/officemd_wasm/www/ for drag-and-drop and sample-fixture testing.

Supported Formats

Format	Extension	Markdown	JSON IR	Docling
Word	.docx	yes	yes	yes
Excel	.xlsx	yes	yes	yes
CSV	.csv	yes	yes	-
PowerPoint	.pptx	yes	yes	yes
PDF	.pdf	yes	yes	-

Development

cargo nextest run --workspace
cargo clippy --workspace --all-targets --exclude officemd_pdf -- -D warnings

For JS and Python tests, see examples/README.md.

Acknowledgements

PDF extraction vendors pdf-inspector by Firecrawl (MIT).

PDF primitives lopdf by J-F-Liu (MIT).

License

Apache 2.0 - see LICENSE.