pylopdf

Edit, render, extract, and create PDFs in Python.

pylopdf aims to give Python programs complete control of PDF: every feature of the PDF 2.0 specification read, interpreted, created, edited, and preserved, including everything that pymupdf does with PDF files, through a smaller API of its own. The roadmap defines the scope, exclusions, and evidence required for a completeness claim; API stability at v1.0 alone will not establish complete coverage. It is built on the Rust PDF engines lopdf, hayro, and krilla, released under the MIT license, and installs as a single wheel with no required Python dependencies.

Documentation · Getting started · API reference · Migrating from pymupdf

Status: pylopdf is pre-1.0. Version 0.16 redesigned its API, as the roadmap describes; the migration guide covers code written for pymupdf or pylopdf 0.15. Complete control is the goal, not yet the state: the roadmap's assessment lists the gaps. The stability policy explains what may change.

Install

pip install pylopdf

Python 3.11 or newer. Wheels are available for Linux, macOS, and Windows, including free-threaded Python 3.14t and 3.15t.

Optional extras add fonts and OCR models:

Extra	Adds
jp, ko, zh-cn, zh-tw	Japanese, Korean, and Chinese fonts for displaying and writing text
ar, he, hi, th	Arabic, Hebrew, Hindi, and Thai fonts for writing text
ocr	An offline OCR engine for scanned pages

pip install "pylopdf[jp,ocr]"

Quickstart

import pylopdf

doc = pylopdf.open("input.pdf")
page = doc.pages[0]

# Extract
text = page.text()                     # plain text
words = page.words()                   # words with their positions
hits = page.search("invoice")          # rectangles of each match
tables = page.tables()                 # tables with their cell text
markdown = doc.markdown()              # whole document as Markdown

# Render
page.pixmap(dpi=150).save("page.png")
svg = page.svg()

# Add content
with page.draw() as canvas:
    canvas.text("Approved", (72, 72), size=20, fill=pylopdf.Color(0.8, 0, 0))
    canvas.image((72, 100, 200, 200), "logo.png")
page.annotations.add(pylopdf.Highlight(hits))

# Reorganize pages (fetch pages again with doc.pages[i] afterwards)
del doc.pages[1:3]
doc.pages.extend(pylopdf.open("appendix.pdf").pages)   # append another document

# Save, compressed and without unused objects
doc.save("output.pdf")

The getting started guide covers each area in more detail.

Features

• Pages and documents: merge, split, reorder, rotate, and delete pages; table of contents, page labels, metadata, and file attachments
• Text extraction: plain or positioned text, search, tables, Markdown output, and vertical Japanese/Chinese text
• Rendering: pages to PNG or SVG, with parallel rendering of many pages
• Adding content: text in any font with proper shaping, images, vector drawings, pages from other PDFs, highlights, and links
• Forms: fill in form fields
• OCR: make scanned pages searchable without installing Tesseract
• Saving: compression, cleanup of unused objects, incremental updates that keep signatures valid, and AES-256 or AES-128 encryption

Why pylopdf

• MIT licensed. pymupdf is AGPL or commercial.
• Small and self-contained. About 7 MB per wheel, with nothing else to install, so it fits AWS Lambda and similar environments. It also runs in WebAssembly, including Cloudflare Python Workers.
• Safe with untrusted files. Size and complexity limits raise a clear LimitError instead of exhausting memory; DocumentLimits.web() enables all of them. See the security model.
• Built for concurrency. Heavy operations release the GIL, and free-threaded Python is supported. See the concurrency guide.

Not supported yet

Redaction, editing or deleting annotations, creating form fields, and validated PDF/A and PDF/UA output are planned; the roadmap lists them.

Typesetting with HTML and CSS and digital signatures stay with ecosystem tools. The ecosystem guide shows how to pair pylopdf with typst and pyHanko for those.

Benchmarks

Benchmarks against pymupdf are reproducible from this repository, and the results include the cases where pylopdf is slower. See the benchmark page.

Contributing

Bug reports and pull requests are welcome. Building from source needs a Rust toolchain and uv:

uv sync
uv run pytest

CONTRIBUTING.md has the full development guide. Please report security issues privately through GitHub Security Advisories.

License

MIT. pylopdf is built on lopdf, hayro, krilla, and HarfRust; see NOTICE.md for third-party licenses.