Skip to content

readme.md

Latest commit

 

History

History
134 lines (96 loc) · 3.07 KB

readme.md

File metadata and controls

134 lines (96 loc) · 3.07 KB

Deadsimple - Document Extractor

FastAPI service for extracting text from PDF, DOCX, PPTX, EPUB, HTML, TXT with multi-backend architecture.

🚀 Features

  • Multiple file formats: Supports PDF, DOCX, PPTX, EPUB, HTML, Markdown, TXT, CSV, and more
  • Vision LLM: Optional OpenAI-compatible API for extracting text from scanned PDFs/images
  • Page-wise output: Returns structured output (one per page/slide)
  • Extensible backend architecture: Designed for easy addition of new backends

📦 Installation

pip install -e .

Supported Formats

markitdown supports these file types out of the box:

Format MIME Type
PDF application/pdf
DOCX application/vnd.openxmlformats-officedocument.wordprocessingml.document
PPTX application/vnd.openxmlformats-officedocument.presentationml.presentation
EPUB application/epub+zip
HTML text/html
Markdown text/markdown or text/x-markdown
TXT text/plain
CSV text/csv
DOC (legacy) application/msword
PPT (legacy) application/vnd.ms-powerpoint
XLSX application/vnd.openxmlformats-officedocument.spreadsheetml.sheet
XLS (legacy) application/vnd.ms-excel

🌐 Vision LLM Mode

Vision mode is configured in backends.ini under the [llm] section:

[llm]
# Set these values to enable vision/OCR mode
token = your_api_key          # OpenAI or compatible API key
model = gpt-4o                # Model name (supports vision)
url = https://api.openai.com/v1  # API endpoint

To enable vision mode:

  1. Open backends.ini
  2. Uncomment and set the [llm] section values
  3. Restart the server

Example with local Ollama:

[llm]
token = ollama
model = llava
url = http://localhost:11434/v1

Leave [llm] empty to disable vision mode (uses markitdown without LLM).

🧪 Running

python deadsimple/main.py

API available at http://localhost:5000

📡 API

Process Document

curl -X POST http://localhost:5000/process \
  -H "Content-Type: application/pdf" \
  --data-binary @document.pdf

Response:

{
  "page_content": "page 1 text\n---\npagina 2 text",
  "metadata": {}
}

🧪 Testing

pytest tests/

Optional Backends

The service supports optional backends that provide extra capabilities such as OCR, document linking, and content analysis. These backends are not required for basic operation; markitdown is always included.

Available backends

  • ocrflux – OCR extraction backend
  • doclings – Document linking backend
  • docstrange – Specialized document processing backend
  • marker – Marker based backend

Installation

Each backend is provided as an extra in pyproject.toml. Install the desired backend with pip, for example:

pip install -e ".[ocrflux]"
pip install -e ".[doclings]"
pip install -e ".[docstrange]"
pip install -e ".[marker]"

You can also install multiple extras at once:

pip install -e ".[ocrflux,doclings,docstrange,marker]"