docxlite

docxlite is a small layer on top of python-docx for reading, previewing, searching, and editing Word documents from Python. It gives .docx files a notebook-friendly markdown preview, adds simple markdown insertion for paragraphs and table cells, and supports tracked insertions and deletions with author attribution.

It is designed for document automation workflows where an agent or notebook needs to make precise edits to Word files without losing the structure that matters: paragraphs, tables, runs, formatting, alignment, merged cells, and revision metadata.

Installation

Install latest from pypi

$ pip install docxlite

How to use

from docxlite import *

Create or open a Word document using DocxDocument. In a notebook or solveit dialog, the document renders as markdown so you can inspect its contents without downloading and opening Word.

doc = DocxDocument()
p = doc.add_paragraph('Hello from docxlite')
doc

Hello from docxlite

add_md parses a small markdown subset and adds formatted runs. DocxDocument.add_paragraph() uses the same markdown path, so you can create formatted paragraphs directly. It supports bold, italic, bold-italic, and underline.

p = doc.add_paragraph('This has **bold**, *italic*, and <u>underline</u>.')
p

This has bold, italic, and underline.

Tables render as markdown tables. add_row accepts cell values directly, and table cells support add_md too. If you pass fewer values than the table has columns, the last provided value is merged across the remaining cells.

tbl = doc.add_table(rows=0, cols=3)
tbl.add_row('Name', 'Value', 'Notes')
tbl.add_row('Apples', '42', '**crisp**')
tbl.add_row('Merged footer')
tbl

Name	Value	Notes
Apples	42	crisp
Merged footer

Documents are iterable in body order. That matters because python-docx exposes doc.paragraphs and doc.tables separately, which loses the interleaving of paragraphs and tables in the original file.

for block in doc:
    print(type(block).__name__, block.text[:40])

Paragraph Hello from docxlite
Paragraph This has bold, italic, and underline.
Table Name  Value   Notes
Apples  42  crisp
Merged 

search returns the matching document blocks themselves, not detached search results. You can search for text, then edit the returned paragraph or table directly.

hits = doc.search('bold')
hits[0]

This has bold, italic, and underline.

Formatting and alignment

apply_base copies paragraph-level formatting and base run defaults from an existing paragraph. Use markdown for explicit emphasis such as bold, italic, and underline; use apply_base when you want a new paragraph to inherit local Word formatting.

base = doc.search('This has')[0]
new_heading = doc.add_paragraph('<u>**A formatted heading**</u>')
new_heading.apply_base(base)

A formatted heading

docxlite.skill also exports the common python-docx alignment enums, so agents can set paragraph, cell, and table alignment without reaching through undocumented constants.

new_heading.alignment = WD_ALIGN_PARAGRAPH.CENTER
new_heading

A formatted heading

Tracked changes

Call set_tracking(author) to make insertions and deletions appear as Word tracked changes. Use set_tracking(None) when you want edits to be applied directly. The LLM skill defaults to Docxlite as the tracked-change author.

set_tracking('Docxlite')
hits[0].insert_after('Inserted with tracking enabled.')
doc

Hello from docxlite

This has bold, italic, and underline.

Inserted with tracking enabled.

Name	Value	Notes
Apples	42	crisp
Merged footer

A formatted heading

Deleting with tracking enabled marks content as deleted instead of removing it. Deleted text appears in the markdown preview using revision spans, and Word will show it as a tracked deletion.

doc.search('Hello')[0].delete()
doc

Hello from docxlite

This has bold, italic, and underline.

Inserted with tracking enabled.

Name	Value	Notes
Apples	42	crisp
Merged footer

A formatted heading

LLM Integration

docxlite ships with a pyskills integration for LLM agents. Importing docxlite.skill exposes the document classes, common Word alignment enums, and editing methods to the tool system. It also enables tracked changes with set_tracking('Docxlite') by default so agent edits are reviewable in Word.

from docxlite.skill import *

The skill uses the same API as regular docxlite; the difference is that the relevant classes and methods are registered for safe tool access. That lets an LLM search, preview, insert, delete, align, merge, and save .docx files while preserving revision metadata.

skill_doc = DocxDocument()
skill_doc.add_paragraph('Inserted through the docxlite pyskill')
skill_doc

Inserted through the docxlite pyskill

Saving

Save the edited document with the normal python-docx API.

doc.save('example.docx')