doctr-index 0.1.2

Open-source document indexing library for building hierarchical trees from PDF and Markdown.

doctr

Deterministic multi-format document tree indexing for RAG and agent workflows.

Overview

doctr turns real-world documents into PageIndex-style tree nodes:

• title
• node_id
• start_index
• end_index
• summary
• nodes

Supported input formats:

• pdf
• docx
• xlsx / xlsm
• md / markdown
• txt
• msg (text fallback)
• embedded files inside supported containers (best effort)

Package Names

• PyPI distribution: doctr-index
• Python import: doctr
• Legacy compatibility import: pdfindexing (shim)

Installation

pip install doctr-index

Optional extras:

pip install 'doctr-index[office]'   # docx/xlsx
pip install 'doctr-index[docling]'  # docling adapter
pip install 'doctr-index[ocr]'      # OCR provider adapter

Quick Start

from doctr import index_document

idx = index_document(
    "/path/to/report.pdf",
    prefer_toc_hierarchy=True,
    include_embedded=True,
    max_embedded_depth=2,
)

tree = idx.to_pageindex_dict(include_empty_nodes=False)
print(tree["nodes"][0])

4-Stage Pipeline

from doctr import DocumentPipeline

pipeline = DocumentPipeline()

# 1) Document input
info = pipeline.document_input("/path/to/report.pdf")

# 2) Docling conversion
converted = pipeline.docling_conversion("/path/to/report.pdf")

# 3) Tree index builder
index = pipeline.build_tree_index(converted=converted)

# 4) Retrieval/chat context
context = pipeline.retrieve_for_chat(index, "What are the key risks?", top_k=6)
print(context)

API Usage Reference

Core indexing

from doctr import (
    index_document,
    index_pdf_file,
    index_docx_file,
    index_xlsx_file,
    index_markdown_file,
    index_markdown_text,
)

idx1 = index_document("/path/to/file.pdf")
idx2 = index_pdf_file("/path/to/file.pdf")
idx3 = index_docx_file("/path/to/file.docx")
idx4 = index_xlsx_file("/path/to/file.xlsx")
idx5 = index_markdown_file("/path/to/file.md")
idx6 = index_markdown_text("# Root\n## Child")

Class API

from doctr import DocumentIndexer

indexer = DocumentIndexer()
idx = indexer.index_document("/path/to/file.pdf", include_embedded=True)

OCR payload mapping (no API key required)

from doctr import DocumentIndexer, document_index_from_ocr_payload

payload = {
    "doc_id": "pi-abc123",
    "status": "completed",
    "result": [
        {
            "title": "Financial Stability",
            "page_index": 21,
            "text": "...",
            "nodes": []
        }
    ]
}

idx = DocumentIndexer().index_with_ocr("/path/to/scanned.pdf", ocr_payload=payload)
idx2 = document_index_from_ocr_payload(payload, source="/path/to/scanned.pdf")

OCR provider adapter (API key required)

from doctr import PageIndexOCRProvider, DocumentIndexer

provider = PageIndexOCRProvider(api_key="YOUR_KEY", base_url="https://api.pageindex.ai")
idx = DocumentIndexer(ocr_provider=provider).index_with_ocr("/path/to/scanned.pdf")

Retrieval helper

from doctr import index_document, retrieve_context

idx = index_document("/path/to/report.pdf")
payload = idx.to_pageindex_dict(include_empty_nodes=False)
ctx = retrieve_context(payload, "What changed in supervision?", top_k=8)
print(ctx)

Custom enricher

from doctr import index_document

def enrich(idx):
    for node in idx.nodes:
        if node.summary:
            node.summary = node.summary[:300]
    return idx

idx = index_document("/path/to/file.pdf", enricher=enrich)

CLI

doctr /path/to/file.pdf \
  --prefer-toc-hierarchy \
  --include-embedded \
  --max-embedded-depth 2 \
  --format pageindex \
  --output output_index.json

Embedded Files

Enable recursive embedded indexing:

from doctr import index_document

idx = index_document("/path/to/container.docx", include_embedded=True, max_embedded_depth=2)

Output includes:

• Embedded Files branch in tree
• metadata["embedded_files"] manifest with indexed/skipped status

Example Scripts

• examples/chat_with_document.py
• examples/docling_pipeline_demo.py
• examples/local_ocr_tree_indexer.py

Notes

• Normal indexing is local and does not require API keys.
• OCR provider mode requires credentials only if using remote OCR.
• nodes: [] means a valid leaf node.

Development

pytest -q

License

MIT