langchain-kreuzberg

Overview

langchain-kreuzberg is a LangChain document loader that wraps Kreuzberg's extraction API. It supports 88+ file formats out of the box, provides true async extraction powered by Rust's tokio runtime, and produces LangChain Document objects enriched with rich metadata including detected languages, quality scores, and extracted keywords.

Installation

pip install langchain-kreuzberg

Requires Python 3.10+.

Quick Start

from langchain_kreuzberg import KreuzbergLoader

loader = KreuzbergLoader(file_path="report.pdf")
docs = loader.load()

print(docs[0].page_content[:200])
print(docs[0].metadata["source"])

Features

• 88+ file formats -- PDF, DOCX, PPTX, XLSX, images, HTML, Markdown, plain text, and many more
• True async -- native async extraction backed by Rust's tokio runtime; no thread-pool workarounds
• Rich metadata -- title, author, page count, detected languages, quality score, extracted keywords, and more
• OCR with 3 backends -- Tesseract, EasyOCR, and PaddleOCR with configurable language support
• Per-page splitting -- yield one Document per page for fine-grained RAG pipelines
• Bytes input -- load documents directly from raw bytes (e.g., API responses, S3 objects)
• Output format selection -- choose between plain text, Markdown, Djot, HTML, or structured output

Usage Examples

Load a PDF with defaults

from langchain_kreuzberg import KreuzbergLoader

loader = KreuzbergLoader(file_path="contract.pdf")
docs = loader.load()

Load multiple files

loader = KreuzbergLoader(
    file_path=["report.pdf", "notes.docx", "data.xlsx"],
)
docs = loader.load()

OCR a scanned document with Tesseract

from kreuzberg import ExtractionConfig, OcrConfig

config = ExtractionConfig(
    force_ocr=True,
    ocr=OcrConfig(backend="tesseract", language="eng"),
)

loader = KreuzbergLoader(
    file_path="scanned.pdf",
    config=config,
)
docs = loader.load()

Load all files from a directory

loader = KreuzbergLoader(
    file_path="./documents/",
    glob="**/*.pdf",
)
docs = loader.load()

Per-page splitting for RAG

from kreuzberg import ExtractionConfig, PageConfig

config = ExtractionConfig(pages=PageConfig(extract_pages=True))

loader = KreuzbergLoader(
    file_path="handbook.pdf",
    config=config,
)
docs = loader.load()
# docs[0].metadata["page"] == 0  (zero-indexed)

Load from bytes (API response)

import httpx

response = httpx.get("https://example.com/report.pdf")

loader = KreuzbergLoader(
    data=response.content,
    mime_type="application/pdf",
)
docs = loader.load()

Advanced config

from kreuzberg import ExtractionConfig, OcrConfig, PageConfig

config = ExtractionConfig(
    output_format="markdown",
    ocr=OcrConfig(backend="easyocr", language="deu"),
    force_ocr=True,
    pages=PageConfig(extract_pages=True),
)

loader = KreuzbergLoader(
    file_path="report.pdf",
    config=config,
)
docs = loader.load()

Async loading

import asyncio
from langchain_kreuzberg import KreuzbergLoader

async def main():
    loader = KreuzbergLoader(file_path="report.pdf")
    docs = await loader.aload()
    print(f"Loaded {len(docs)} documents")

asyncio.run(main())

API Reference

KreuzbergLoader

from langchain_kreuzberg import KreuzbergLoader

Extends langchain_core.document_loaders.BaseLoader.

Constructor Parameters

All parameters are keyword-only.

Methods

Metadata Fields

Each Document produced by KreuzbergLoader includes the following metadata fields (when available):

Additional metadata fields from Kreuzberg's document-level metadata are flattened into the metadata dict when present.

Supported Formats

Kreuzberg supports 88+ file formats including PDF, DOCX, images (via OCR), spreadsheets, presentations, HTML, Markdown, and many more. For the full and up-to-date list of supported formats, see the Kreuzberg repository.

Contributing

This project uses uv for dependency management.

# Clone the repository
git clone https://github.com/kreuzberg-dev/langchain-kreuzberg.git
cd langchain-kreuzberg

# Install dependencies (including dev group)
uv sync

# Run linting
uv run ruff check .
uv run ruff format --check .
uv run mypy .

# Run unit tests
uv run pytest --cov

# Run integration tests (real file extraction, no mocks)
uv run pytest -m integration -v

# Install pre-commit hooks
prek install

License

This project is licensed under the MIT License.