langchain-hwp-hwpx-loader 0.1.0

LangChain document loaders for HWP/HWPX using hwp-hwpx-parser

LangChain HWP/HWPX Loader

langchain-hwp-hwpx-loader is a pure-Python LangChain loader for Korean .hwp and .hwpx documents, powered by hwp-hwpx-parser.

It is built for offline/on-prem RAG indexing and keeps body text, tables, footnotes/endnotes, memos, and hyperlinks with configurable policies.

Installation

pip install langchain-hwp-hwpx-loader

Quickstart

Single document mode

from pathlib import Path

from hwp_hwpx_parser import ExtractOptions, ImageMarkerStyle, TableStyle
from langchain_hwp_hwpx import HwpHwpxLoader

options = ExtractOptions(
    table_style=TableStyle.MARKDOWN,
    image_marker=ImageMarkerStyle.SIMPLE,
)

loader = HwpHwpxLoader(
    file_path=Path("docs/sample.hwp"),
    mode="single",
    extract_options=options,
    include_tables=True,
    include_notes=True,
    include_memos=True,
    include_hyperlinks=True,
)

docs = loader.load()
print(docs[0].page_content[:400])
print(docs[0].metadata)

Elements mode

from langchain_hwp_hwpx import HwpHwpxLoader

loader = HwpHwpxLoader("docs/sample.hwpx", mode="elements")
for doc in loader.lazy_load():
    print(doc.metadata["element_type"], doc.metadata["element_index"])

Directory loading

from langchain_hwp_hwpx import HwpHwpxDirectoryLoader

directory_loader = HwpHwpxDirectoryLoader(
    dir_path="docs",
    glob="**/*",
    recursive=True,
    mode="single",
    on_error="warn",
)

docs = directory_loader.load()
print(len(docs))

Main Options

• mode: "single" or "elements"
• include_tables, include_notes, include_memos, include_hyperlinks
• include_images, images_dir, image_document_mode
• on_encrypted: "raise" | "skip" | "placeholder"
• on_invalid: "raise" | "skip" | "placeholder"
• on_error: "raise" | "skip" | "warn"

extract_options accepts hwp_hwpx_parser.ExtractOptions.

Metadata

All returned documents include common metadata:

• source, file_name, file_type
• loader, parser
• extracted_at (UTC ISO timestamp by default)

mode="elements" adds:

• element_type, element_index
• table fields: row_count, col_count
• note fields: note_type, note_number
• optional memo/hyperlink/image specific fields when available

Limitations

• Encrypted documents are only detectable. Decryption is not supported.
• OCR and visual layout reconstruction are out of scope.
• Parsing quality depends on hwp-hwpx-parser capabilities.

Compatibility

• Python: >=3.10,<4.0
• langchain-core>=1.0.0,<2.0.0
• hwp-hwpx-parser>=1.0.0,<2.0.0