Kreuzberg is a library for simplified text extraction from PDF files. It's meant to offer simple, hassle free text extraction.

Why?

I am building, like many do now, a RAG focused service (checkout https://grantflow.ai). I have text extraction needs. There are quite a lot of commercial options out there, and several open-source + paid options. But I wanted something simple, which does not require expansive round-trips to an external API. Furthermore, I wanted something that is easy to run locally and isn't very heavy / requires a GPU.

Hence, this library.

• Extract text from PDFs, images, office documents and more (see supported formats below)
• Use modern Python with async (via anyio) and proper type hints
• Extensive error handling for easy debugging

1. Begin by installing the python package:

   pip install kreuzberg
   

2. Install the system dependencies:

• pandoc (non-pdf text extraction, GPL v2.0 licensed but used via CLI only)
• tesseract-ocr (for image/PDF OCR, Apache License)

This library is built to be minimalist and simple. It also aims to utilize OSS tools for the job. Its fundamentally a high order async abstraction on top of other tools, think of it like the library you would bake in your code base, but polished and well maintained.

• PDFs are processed using pdfium2 for searchable PDFs + Tesseract OCR for scanned documents
• Images are processed using Tesseract OCR
• Office documents and other formats are processed using Pandoc
• PPTX files are converted using python-pptx
• HTML files are converted using html-to-markdown
• Plain text files are read directly with appropriate encoding detection

V1:

• - html file text extraction
• - better PDF table extraction
• - TBD

V2:

• - extra install groups (to make dependencies optional)
• - metadata extraction (possible breaking change)
• - TBD

Feel free to open a discussion in GitHub or an issue if you have any feature requests

Is welcome! Read guidelines below.

Kreuzberg supports a wide range of file formats:

• PDF (.pdf) - both searchable and scanned documents
• Word Documents (.docx, .doc)
• Power Point Presentations (.pptx)
• OpenDocument Text (.odt)
• Rich Text Format (.rtf)

• JPEG, JPG (.jpg, .jpeg, .pjpeg)
• PNG (.png)
• TIFF (.tiff, .tif)
• BMP (.bmp)
• GIF (.gif)
• WebP (.webp)
• JPEG 2000 (.jp2, .jpx, .jpm, .mj2)
• Portable Anymap (.pnm)
• Portable Bitmap (.pbm)
• Portable Graymap (.pgm)
• Portable Pixmap (.ppm)

• HTML (.html, .htm)
• Plain Text (.txt)
• Markdown (.md)
• reStructuredText (.rst)
• LaTeX (.tex)

• Comma-Separated Values (.csv)
• Tab-Separated Values (.tsv)

Kreuzberg exports two async functions:

• Extract text from a file (string path or pathlib.Path) using extract_file()
• Extract text from a byte-string using extract_bytes()

from pathlib import Path
from kreuzberg import extract_file


# Extract text from a PDF file
async def extract_pdf():
    result = await extract_file("document.pdf")
    print(f"Extracted text: {result.content}")
    print(f"Output mime type: {result.mime_type}")


# Extract text from an image
async def extract_image():
    result = await extract_file("scan.png")
    print(f"Extracted text: {result.content}")


# or use Path

async def extract_pdf():
    result = await extract_file(Path("document.pdf"))
    print(f"Extracted text: {result.content}")
    print(f"Output mime type: {result.mime_type}")

from kreuzberg import extract_bytes


# Extract text from PDF bytes
async def process_uploaded_pdf(pdf_content: bytes):
    result = await extract_bytes(pdf_content, mime_type="application/pdf")
    return result.content


# Extract text from image bytes
async def process_uploaded_image(image_content: bytes):
    result = await extract_bytes(image_content, mime_type="image/jpeg")
    return result.content

When extracting a PDF file or bytes, you might want to force OCR - for example, if the PDF includes images that have text that should be extracted etc. You can do this by passing force_ocr=True:

from kreuzberg import extract_bytes


# Extract text from PDF bytes and force OCR
async def process_uploaded_pdf(pdf_content: bytes):
    result = await extract_bytes(pdf_content, mime_type="application/pdf", force_ocr=True)
    return result.content

Kreuzberg raises two exception types:

Raised when there are issues with input validation:

• Unsupported mime types
• Undetectable mime types
• Path doesn't point at an exist file

Raised when there are issues during the text extraction process:

• PDF parsing failures
• OCR errors
• Pandoc conversion errors

from kreuzberg import extract_file
from kreuzberg.exceptions import ValidationError, ParsingError


async def safe_extract():
    try:
        result = await extract_file("document.doc")
        return result.content
    except ValidationError as e:
        print(f"Validation error: {e.message}")
        print(f"Context: {e.context}")
    except ParsingError as e:
        print(f"Parsing error: {e.message}")
        print(f"Context: {e.context}")  # Contains detailed error information

Both error types include helpful context information for debugging:

try:
    result = await extract_file("scanned.pdf")
except ParsingError as e:
# e.context might contain:
# {
#    "file_path": "scanned.pdf",
#    "error": "Tesseract OCR failed: Unable to process image"
# }

All extraction functions return an ExtractionResult named tuple containing:

• content: The extracted text as a string
• mime_type: The mime type of the output (either "text/plain" or, if pandoc is used- "text/markdown")

from kreuzberg import ExtractionResult


async def process_document(path: str) -> str:
    result: ExtractionResult = await extract_file(path)
    return result.content


# or access the result as tuple

async def process_document(path: str) -> str:
    content, mime_type = await extract_file(path)
    # do something with mime_type
    return content

This library is open to contribution. Feel free to open issues or submit PRs. Its better to discuss issues before submitting PRs to avoid disappointment.

1. Clone the repo
2. Install the system dependencies
3. Install the full dependencies with uv sync
4. Install the pre-commit hooks with:

   pre-commit install && pre-commit install --hook-type commit-msg

5. Make your changes and submit a PR

This library uses the MIT license.