Quickstart

Extract a document and explicitly create chunks for retrieval.

1. Install

pip install longparser

2. Parse a PDF

from longparser import ChunkingConfig, DocumentPipeline, ProcessingConfig

# Create pipeline with defaults
config = ProcessingConfig()
pipeline = DocumentPipeline(config)

# Parse a PDF
result = pipeline.process_file("research_paper.pdf", config=config)
chunks = pipeline.chunk(result, ChunkingConfig())

print(f"Pages: {result.document.metadata.total_pages}")
print(f"Chunks: {len(result.chunks)}")
if chunks:
    print(chunks[0].text)

process_file() extracts blocks and hierarchy; chunk() fills result.chunks. Embedding, indexing and retrieval are separate stages. The default PDF OCR path requires a Tesseract executable and appropriate language data. For documents that do not need OCR, choose processing options appropriate to the input.

3. Inspect Chunks

for chunk in result.chunks[:3]:
    print(f"[{chunk.chunk_type}] tokens={chunk.token_count}")
    print(chunk.text[:200])
    print("---")

4. Use with LangChain

Install longparser[langchain] to use the loader.

from longparser.integrations.langchain import LongParserLoader

loader = LongParserLoader("report.pdf")
documents = loader.load()  # Returns List[Document]

5. Use with LlamaIndex

Install longparser[llamaindex] to use the reader.

from longparser.integrations.llamaindex import LongParserReader

reader = LongParserReader()
nodes = reader.load_data(file="report.pdf")

6. Start the REST Server

# Set environment variables
cp .env.example .env
# Edit .env with your keys

# Start the API, background worker, MongoDB and Redis
docker compose up -d --build

Then visit http://localhost:8000/docs for the Swagger UI.

Supported Formats

Format	Extension	Notes
PDF	.pdf	OCR + table structure
Word	.docx	OMML equation injection
PowerPoint	.pptx	Slide-by-slide chunking
Excel	.xlsx	Sheet-aware table parsing
CSV	.csv	Column-profile chunks