langchain-docdigitizer 0.2.0

LangChain document loader for the DocDigitizer document processing API

langchain-docdigitizer

LangChain document loader for the DocDigitizer document processing API.

v0.1.x is deprecated. Upgrade to v0.2.0+ for the new API endpoint. The previous endpoint (https://apix.docdigitizer.com/sync) will be removed in a future release.

Installation

pip install langchain-docdigitizer

Usage

from langchain_docdigitizer import DocDigitizerLoader

# Load a single PDF
loader = DocDigitizerLoader(api_key="dd_live_...")
docs = loader.load("invoice.pdf")

print(docs[0].page_content)       # JSON with extracted fields
print(docs[0].metadata)           # document_type, confidence, country_code, etc.

# Load all PDFs from a directory
loader = DocDigitizerLoader(api_key="dd_live_...", file_path="invoices/")
docs = loader.load()

# Use in a RAG pipeline
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain_openai import OpenAIEmbeddings
from langchain_community.vectorstores import FAISS

splitter = RecursiveCharacterTextSplitter(chunk_size=1000)
chunks = splitter.split_documents(docs)
vectorstore = FAISS.from_documents(chunks, OpenAIEmbeddings())

Configuration

Parameter	Environment Variable	Default
api_key	DOCDIGITIZER_API_KEY	—
base_url	DOCDIGITIZER_BASE_URL	https://api.docdigitizer.com/v3/docingester
timeout	DOCDIGITIZER_TIMEOUT	300
max_retries	—	3
pipeline	—	None
content_format	—	"json"

Content Formats

• "json" (default): page_content is a JSON string of extracted fields
• "text": page_content is key-value pairs separated by newlines (key: value)
• "kv": page_content is key=value pairs separated by newlines

Document Metadata

Each LangChain Document includes metadata:

Field	Type	Description
source	str	File path of the processed PDF
document_type	str	Detected document type (e.g., "Invoice")
confidence	float	Classification confidence (0-1)
country_code	str	Detected country code (e.g., "PT")
pages	list[int]	Page numbers where document was found
page_range	dict	Start/end page range
trace_id	str	Unique trace identifier

License

MIT