docrouter 0.1.0

Unified document interrogation: retrieve, screenshot, search

docrouter

A swiss army knife for document handling in LLM agents and apps.

docrouter provides a unified API to extract text, search content, and render pages across multiple document formats. It's designed for rapid prototyping - when you need to quickly build an agent that can work with PDFs, Word docs, PowerPoints, and more without wrestling with a dozen different libraries.

What this is

• A single interface for text extraction and search across PDF, DOCX, PPTX, HTML, Markdown, and plain text
• Built with LLM tool-calling in mind (includes a ready-to-use tool API with document registry)
• Zero core dependencies - install only what you need
• Good enough extraction for most prototyping and simple production use cases

What this isn't

This library doesn't compete with state-of-the-art (often paid, API-based) solutions for individual file types. If you need production-grade PDF extraction with perfect table parsing, OCR, or layout analysis, check out specialized tools like Datalab - they're excellent at what they do.

docrouter is for when you need to get something working quickly across multiple formats without overthinking it.

Installation

pip install docrouter[all]  # all formats
pip install docrouter[pdf]  # PDF only
pip install docrouter       # txt/md only (no deps)

Optional dependencies by format:

• pdf - PyMuPDF
• docx - python-docx
• pptx - python-pptx
• html - BeautifulSoup4, lxml

Quick start

from docrouter import open_document

doc = open_document("report.pdf")
doc.info()              # metadata: pages, title, etc.
doc.get_text()          # full document text
doc.search("revenue")   # find text with context
doc.render_page(0)      # screenshot page as PNG (PDF only)

Tool API (for LLM agents)

The tool API maintains a document registry, making it easy to integrate with function-calling LLMs:

from docrouter.tools import (
    open_document_tool,
    search_tool,
    get_text_tool,
    close_document_tool
)

# Open and register a document
result = open_document_tool("quarterly_report.pdf")
doc_id = result["document_id"]

# Search across the document
hits = search_tool(doc_id, "operating income", max_hits=5)

# Get full text or specific pages
text = get_text_tool(doc_id)

# Clean up when done
close_document_tool(doc_id)

Supported formats

Format	Extensions	Unit type	Features
PDF	.pdf	page	text, search, render
Word	.docx	section	text, search, tables
PowerPoint	.pptx	slide	text, search, tables, notes
HTML	.html, .htm	section	text, search
Markdown	.md	chunk	text, search
Plain text	.txt, .csv, .json, etc.	chunk	text, search
Code	.py, .js, .ts, etc.	chunk	text, search
Images	.png, .jpg, etc.	-	metadata only

Contributing

Contributions are welcome! If you'd like to extend docrouter with new formats, better extraction for existing ones, or other improvements, please open an issue or PR.

License

MIT