Installation

Requirements

Python 3.10 or later
pip 22+ (run pip install --upgrade pip if unsure)

Base install

pip install docslicer

This covers PDF, DOCX, and PPTX parsing out of the box. No system dependencies required.

DocSlicer's own wheel is small — the bulk of the install comes from its dependencies:

Package	Download (wheel)	Installed (on disk)	Notes
`docslicer`	~0.6 MB	~4 MB	The library itself
`pandas`	~9.5 MB	~66 MB	DataFrames flow through every pipeline step
`numpy`	~5.1 MB	~31 MB	Coordinate math and chunk partitioning
`lxml`	~8.2 MB	~19 MB	XML tree parser for DOCX and PPTX
`pypdfium2`	~3.5 MB	~8 MB	PDF text, images, shapes, links, and metadata extraction
`pikepdf`	~4.5 MB	~13 MB	PDF parsing, decryption, and page-level recovery
`Pillow`	~4.8 MB	~15 MB	Image handling; pulled in by `pikepdf`
`langdetect`	~1.0 MB	~2 MB	Detects the language of extracted text
`pyyaml`	~0.2 MB	~1 MB	YAML config parsing
`beautifulsoup4`	~0.1 MB	~1 MB	HTML parsing and static-page fallback
`httpx`	~0.1 MB	~1 MB	HTTP fallback when Playwright unavailable
Total	~38 MB	~161 MB

Measured on macOS arm64. The lxml wheel is larger on macOS because it ships a universal2 (arm64 + x86_64) binary; Linux builds are ~4 MB downloaded / ~8 MB installed.

Optional extras

Extra	Package	Download	Installed (on disk)	Notes
html	`playwright`	~4 MB	~15 MB	Browser automation driver
html	`playwright install chromium`	~150 MB	~150 MB	Browser binary; not a pip package
llm	`tiktoken`	~1 MB	~3 MB	Exact token counts (cl100k_base); chars/4 fallback
ocr	`tesserocr`	~3.8 MB	~9.5 MB	Direct `libtesseract` bindings (bundled in the wheel), plus `cysignals`
ocr	`brew install tesseract`	~20 MB	~35 MB	Tesseract engine + language models; not a pip package
ocr	`opencv-python`	~48 MB	~119 MB	Image pre-processing before OCR
ocr	`Pillow`	—	—	Already installed with the base install (via `pikepdf`)
crypto	`msoffcrypto-tool`	~4 MB	~14 MB	Password-protected DOCX/PPTX; pulls in `cryptography`
parquet	`pyarrow`	~25 MB	~90 MB	Parquet read/write for chunk export
Total (all extras)		~256 MB	~436 MB

HTML parsing

To parse URLs or JS-rendered pages, install the html extra:

pip install 'docslicer[html]'

Then install the Chromium browser that Playwright will drive:

playwright install chromium

Only Chromium is needed — DocSlicer does not use Firefox or WebKit.

Playwright is strongly recommended. DocSlicer will not crash without it — plain HTML fetching via httpx still works — but output quality degrades significantly. HTML structure detection relies on styling cues (bold, italic, underline, font size, color, ...) to distinguish headings from paragraphs and build an accurate document hierarchy. Static HTML pages carry almost none of this information. Skip Playwright only if you know your target pages are well-structured semantic HTML.

Token counting

The llm extra adds tiktoken for accurate token-count estimates on chunks:

pip install 'docslicer[llm]'

OCR

The ocr extra enables text extraction from scanned PDFs via Tesseract:

pip install 'docslicer[ocr]'

DocSlicer binds libtesseract directly through tesserocr. The tesserocr wheel bundles the library itself but not the language models, so the Tesseract engine still has to be installed on your system — DocSlicer auto-detects the path:

# macOS
brew install tesseract
 
# Debian / Ubuntu
apt install tesseract-ocr

If pip has to build tesserocr from source (no wheel for your platform), install the development libraries first:

# macOS
brew install tesseract leptonica
 
# Debian / Ubuntu
apt install tesseract-ocr libtesseract-dev libleptonica-dev pkg-config

OCR is a fallback, not a primary use case. DocSlicer is built for native PDFs — documents with real embedded text and font data — and excels at extracting accurate structure from them. OCR works by rendering pages to images and reading pixels, which loses the precise character coordinates, font metrics, and styling cues that drive DocSlicer's hierarchy detection. If your PDFs are natively digital, skip this extra entirely.

Password-protected Office files

The crypto extra adds msoffcrypto-tool, which decrypts password-protected .docx and .pptx files before parsing:

pip install 'docslicer[crypto]'

Pass the password with the password parameter — see parse_document. Encrypted PDFs are handled by pikepdf in the base install and do not need this extra.

Parquet export

pip install 'docslicer[parquet]'

Everything at once

pip install 'docslicer[html,llm,ocr,crypto,parquet]'
playwright install chromium

Playwright setup

Playwright requires a one-time browser download. The playwright install chromium command stores the browser in a local cache (~/.cache/ms-playwright on Linux/macOS).

CI environments — add the install step before your test run:

- run: pip install 'docslicer[html]'
- run: playwright install chromium

Docker — install the system dependencies Playwright needs:

playwright install-deps chromium
playwright install chromium

Version pinning

Pin the exact version in requirements.txt:

docslicer==0.2.0

Or with extras:

docslicer[html]==0.2.0

In pyproject.toml:

[project]
dependencies = [
    "docslicer==0.2.0",
]

To check the latest version:

pip index versions docslicer

Upgrading

pip install --upgrade docslicer

After upgrading, re-run playwright install chromium if you use the html extra — Playwright occasionally requires a newer browser binary.

Next steps

Quickstart — parse your first document in 2 minutes
Format parsers — full parameter reference for all parsers