NEW: Use a Remote DGX as Your Local Vision LLM Provider!
You can connect the extractor to a powerful remote DGX (or any server with GPUs) running Ollama and use it as your private LLM Vision backend—directly from your laptop or office PC. This enables blazing-fast, enterprise-grade document extraction with full privacy, as all data stays within your network. Just set the
OLLAMA_BASE_URLto your DGX/server IP, or change it live from the web UI!
A 100% privacy-compliant, local AI-powered document extraction tool. Perfect for public administrations, law firms, and anyone handling sensitive data. It uses Vision LLMs (like Qwen, Llama3.2-Vision, DeepSeek) through Ollama to extract structured data from any PDF or image into Excel, without ever sending your data to the cloud.
- 100% Privacy & Local Executed: No data leaves your machine/network. Essential for GDPR compliance.
- Zero-Shot Extraction: No need to map coordinate templates. The AI "reads" the document naturally.
- Hub & Spoke Architecture Ready: Run Ollama on a single powerful server (Hub) and access the web UI from any office laptop (Spoke) by pointing the UI to your server's IP.
- Smart Structuring: Transform messy layout and loose data points into structured JSON payload.
- Human-in-the-Loop Validation: An elegant side-by-side UI to verify and edit the AI's extraction before exporting to Excel.
- Smart Model Auto-Fallback: Optimized logic for dealing with LLM quirks, markdown block stripping, and empty JSON freezes.
- Install Ollama on your local machine or a server.
- Install Python 3.10+
- Install
poppler-utils(Required for PDF processing)- Ubuntu/Debian:
sudo apt-get install poppler-utils - macOS:
brew install poppler - Windows: Download Poppler for Windows and add to PATH.
- Ubuntu/Debian:
We've provided a simple script to set up everything automatically:
# 1. Clone the repository
git clone https://github.com/your-username/local_data_extractor.git
cd local_data_extractor
# 2. Start the application (Installs deps automatically on first run)
./start.sh
# 3. To gracefully stop the application running in background
./stop.shstart.sh will:
- Create a virtual environment and install dependencies.
- Start the Flask backend.
- Automatically open
http://localhost:5000in your browser.
If you prefer to set up manually or you are on Windows:
# 1. Create & activate virtual environment
python3 -m venv .venv
source .venv/bin/activate # On Windows: .venv\Scripts\activate
# 2. Install dependencies
pip install -r requirements.txt
# 3. Pull a recommended vision model
ollama pull qwen3.5:4b # Great for standard laptops
# ollama pull llama3.2-vision:11b # Great for 12GB+ VRAM
# 4. Start the application
python src/app.pyYou can configure the application using environment variables or a .env file (copy .env.example to .env):
OLLAMA_BASE_URL=http://localhost:11434 # Point this to your remote DGX / Server IP to use a remote GPU as your LLM Vision provider
OLLAMA_MODEL=llama3.2-vision:11b # Default model
PORT=5000 # Web UI portNote: You can also dynamically change the Ollama IP (including a remote DGX or server) and the Model directly from the Web Interface without restarting the application!
local_data_extractor/
├── src/
│ ├── app.py # Flask server and routing
│ ├── processor.py # Core OCR / AI logic, image resizing, and parsing
│ ├── models_config.py # Model catalog and hardware recommendations
│ └── templates/
│ └── index.html # Frontend Vanilla JS / HTML
├── start.sh # One-click start script
├── stop.sh # One-click script to stop the backend
├── setup.sh # Environment setup script
├── requirements.txt # Python dependencies
└── README.md # This file
Distributed under the MIT License. See LICENSE for more information.