runeextract 0.3.0

One extraction API for every document type

RuneExtract

One extraction API for every document type.

RuneExtract is a universal document extraction library that provides a single, consistent API for extracting content from any file type. Whether it's PDF, DOCX, HTML, images, YouTube videos, or Notion exports — RuneExtract returns the same structured output every time.

from runeextract import extract

doc = extract("report.pdf")

print(doc.text)
print(doc.tables)
print(doc.images)
print(doc.metadata)
print(doc.chunks())

One API. Any file. Same output schema.

Installation

pip install runeextract

Optional extras:

Extra	Packages	Feature
[ocr]	easyocr, Pillow	OCR for images and scanned PDFs
[ai]	openai	AI summarization, keywords, entities, Q&A, flashcards
[youtube]	youtube-transcript-api, yt-dlp	YouTube transcript and metadata
[notion]	requests, aiohttp	Notion page and database extraction
[epub]	EbookLib, beautifulsoup4	EPUB e-book extraction
[async]	aiohttp	Async HTTP support for URL extractors
[all]	All of the above	Everything

Quick Start

from runeextract import extract

doc = extract("report.pdf", ocr=True, tables=True, chunking_strategy="semantic")
print(doc.text)
print(doc.tables[0].to_dataframe())

Batch & Async

from runeextract import extract_many, extract_many_async

docs = extract_many(["a.pdf", "b.docx", "c.html"])

docs = await extract_many_async(["a.pdf", "b.docx", "c.html"], max_concurrency=4)

CLI

runeextract file.pdf --ocr --chunking semantic --tree
runeextract https://youtube.com/watch?v=... --youtube-format transcript
runeextract scanned.pdf --ocr --output-dir ./output
runeextract ./docs --watch         # Watch directory for new files

Supported Formats

Format	Status	Content Extracted
PDF	✅ v0.2.0	text, tables, images, metadata, scanned-page OCR
DOCX	✅ v0.2.0	paragraphs, tables, images, metadata, image OCR
PPTX	✅ v0.2.0	slides, tables, images, metadata, speaker notes, image OCR
XLSX	✅ v0.2.0	worksheets, tables, metadata, multiple sheets
HTML	✅ v0.2.0	headings, paragraphs, tables, links, meta tags
Markdown	✅ v0.2.0	headings, lists, code blocks, tables, frontmatter
CSV	✅ v0.2.0	tables, text, row/column metadata
JSON	✅ v0.2.0	pretty-print, table from list-of-dicts
Images (PNG/JPG/TIFF/BMP/WebP)	✅ v0.2.0	metadata (width, height, format), OCR text
EPUB	✅ v0.2.0	text, tables, images, metadata (title, author, etc.)
YouTube	✅ v0.2.0	transcript, timestamps, chapters, metadata
Notion	✅ v0.2.0	pages, databases, 14 block types, async

Features

Intelligent Chunking

chunks = doc.chunks(
    strategy="semantic",  # by_page, by_heading, semantic, fixed_size
    size=1000
)

AI Processing (optional)

doc = extract("report.pdf")
print(doc.summary())
print(doc.keywords())
print(doc.entities())
print(doc.questions())
print(doc.flashcards())

OCR (optional)

doc = extract("invoice.jpg", ocr=True, ocr_lang="en,fr")

Streaming

from runeextract import extract_stream

async for page_doc in extract_stream("large.pdf"):
    process(page_doc)

Plugin System

from runeextract.core.registry import register_extractor

@register_extractor(".txt")
class TxtExtractor(BaseExtractor):
    def extract(self, file_path):
        return Document(text=open(file_path).read(), source_type="txt")

Configuration

export RUNEEXTRACT_OCR=true
export RUNEEXTRACT_MAX_FILE_SIZE=999999999

Or create ~/.runeextract.json:

{"ocr": true, "tables": false, "max_file_size": 1000000}

Caching

from runeextract.core.cache import ExtractionCache
cache = ExtractionCache(ttl=3600)

Project Structure

runeextract/
├── __init__.py          # Public API (extract, extract_many, extract_async, etc.)
├── config.py            # Configuration system (env, JSON, pyproject.toml)
├── exceptions.py        # Custom exception hierarchy
├── cli/main.py          # 14 CLI flags
├── core/
│   ├── extractor.py     # BaseExtractor, StreamingExtractor
│   ├── router.py        # ExtractorRouter (19 file extensions)
│   ├── registry.py      # Plugin registry with entry-point discovery
│   ├── cache.py         # diskcache/JSON cache layer
│   ├── schemas.py       # ExtractionOptions, ExtractionResult
│   └── streaming.py     # get_streaming_extractor, wrapped fallback
├── extractors/
│   ├── pdf/             # PDF (PyMuPDF + pdfplumber)
│   ├── docx/            # DOCX (python-docx)
│   ├── pptx/            # PPTX (python-pptx)
│   ├── xlsx/            # XLSX (openpyxl)
│   ├── html/            # HTML (BeautifulSoup)
│   ├── markdown/        # Markdown (markdown-it-py)
│   ├── csv/             # CSV (stdlib csv)
│   ├── json/            # JSON (stdlib json)
│   ├── image/           # Image (Pillow + easyocr)
│   ├── epub/            # EPUB (EbookLib)
│   ├── youtube/         # YouTube (youtube-transcript-api + yt-dlp)
│   └── notion/          # Notion (REST API + aiohttp)
├── processors/
│   ├── ocr.py           # easyocr-based OCR
│   └── ai.py            # OpenAI / local transformers AI processor
├── models/
│   └── document.py      # Document, Table, Image, Chunk, ChunkingStrategy
└── tests/
    ├── 103 tests across 18 files
    └── benchmarks/

Architecture

File/URL → Router (detects type) → Extractor → Document (unified schema)

Development

pip install -e ".[dev]"
pytest                     # 103 tests

License

MIT — see LICENSE.

Why RuneExtract?

Instead of learning PyMuPDF, python-docx, BeautifulSoup, EasyOCR, etc.:

extract(anything)

That simplicity is the entire product.

---

RuneExtract v0.2.0 — One API to extract them all.