DocSummarizer

A fully offline document summarization tool powered by a local AI model. Designed for scientists and researchers who need to quickly summarize academic papers and documents without sending data to external services.

Features

• 100% Offline: After the initial model download, everything runs locally on your machine
• Privacy-First: Documents never leave your computer - no cloud services, no data collection
• Multiple Formats: Supports PDF, DOCX, DOC, RTF, TXT, and Markdown files
• Batch Processing: Summarize entire folders of documents at once
• Flexible Output: Choose between brief, detailed, or structured summaries
• Cross-Platform: Works on Windows, macOS, and Linux
• Adjustable CPU Usage: Control how many CPU threads to use via Settings

System Requirements

Requirement	Minimum	Recommended
OS	Windows 10, macOS 10.14, Linux	Latest version
RAM	8 GB	16 GB
Storage	6 GB free	10 GB free
CPU	4 cores	8+ cores
Python	3.10+	3.11+

Note: The tool runs on CPU by default. GPU acceleration is optional (see Advanced section).

Quick Start

Option A: Download Standalone Executable (Easiest)

Download the latest release for your platform - no Python required:

1. Go to Releases
2. Download:
   • Windows: DocSummarizer.exe
   • Linux: DocSummarizer
3. Run the executable
4. On first launch, click "Download Model" (~4.4 GB, one-time)

Option B: Run from Source

1. Clone or Download

git clone https://github.com/Wintersta7e/Doc-Summarizer.git
cd Doc-Summarizer

2. Run Setup Script

Windows:

setup_and_run.bat

Linux/macOS:

chmod +x setup_and_run.sh
./setup_and_run.sh

3. First Launch

On first launch, the application will:

1. Create a virtual environment
2. Install dependencies
3. Prompt you to download the AI model (~4.4 GB, one-time)

After setup, the GUI will open automatically.

Usage

Graphical Interface (GUI)

1. Launch with python run.py or use the setup script
2. Click Select File to choose a document
3. Select summary type: Brief, Detailed, or Structured
4. Click Summarize and wait for processing
5. Save the result using Save Summary

Command Line Interface (CLI)

# Summarize a single file
python src/cli.py document.pdf

# Choose summary type
python src/cli.py document.pdf -t structured
python src/cli.py document.pdf -t brief
python src/cli.py document.pdf -t detailed

# Save output to file
python src/cli.py document.pdf -o summary.txt

# Batch process a folder
python src/cli.py ./papers/ -o ./summaries/

# Download model only (no processing)
python src/cli.py --download-only

Summary Types

Type	Description	Best For
Brief	1 paragraph (3-5 sentences)	Quick overview
Detailed	Comprehensive with key points	Understanding content
Structured	Organized sections (Purpose, Methods, Conclusions, etc.)	Academic papers

Project Structure

DocSummarizer/
├── run.py                  # Main entry point (GUI)
├── requirements.txt        # Python dependencies
├── README.md               # This file
├── DEVELOPMENT.md          # Developer documentation
├── DocSummarizer.spec      # PyInstaller build configuration
├── setup_and_run.bat       # Windows launcher
├── setup_and_run.sh        # Linux/macOS launcher
├── .gitignore              # Git ignore rules
└── src/
    ├── __init__.py
    ├── gui.py              # GUI application (CustomTkinter)
    ├── cli.py              # Command-line interface
    ├── document_parser.py  # Document text extraction
    ├── model_manager.py    # LLM download and inference
    └── logger.py           # Logging and diagnostics

How It Works

1. Document Parsing: Extracts text from PDF, DOCX, and other formats using pypdf and python-docx
2. Text Processing: Prepares the extracted text for the AI model
3. Local LLM Inference: Uses llama-cpp-python to run a quantized Mistral 7B model
4. Summary Generation: The model generates a summary based on the selected type

Model Information

Property	Value
Model	Mistral 7B Instruct v0.2
Quantization	Q4_K_M (4-bit)
Size	~4.4 GB
Source	HuggingFace (TheBloke)
Context Window	8192 tokens

The model is downloaded on first launch and stored in:

• Windows: %LOCALAPPDATA%\DocSummarizer\models\
• macOS: ~/Library/Application Support/DocSummarizer/models/
• Linux: ~/.local/share/DocSummarizer/models/

Performance

Document Size	Processing Time (CPU)
Short (1-5 pages)	30-60 seconds
Medium (5-15 pages)	1-2 minutes
Long (15+ pages)	2-3 minutes

Note: Times vary based on CPU and thread settings. By default, uses half of available CPU cores to balance speed and system responsiveness. Adjust in Settings > CPU Threads if needed.

Troubleshooting

Model download fails

• Check internet connection
• Ensure 5+ GB free disk space
• Try running as administrator

Out of memory

• Close other applications
• Ensure at least 8 GB RAM
• Process smaller documents

Slow performance

• Normal on CPU - the model is computationally intensive
• Increase CPU threads in Settings for faster processing
• Close other applications to free resources
• Consider GPU acceleration (see DEVELOPMENT.md)

High CPU usage

• Go to Settings > CPU Threads and lower the thread count
• Using fewer threads reduces CPU load but increases processing time

Checking logs for errors

Log files are stored at:

• Windows: %LOCALAPPDATA%\DocSummarizer\logs\
• Linux: ~/.local/share/DocSummarizer/logs/

Logs contain startup info, performance metrics, and error details (no document content is logged).

PDF extraction issues

• Some scanned PDFs (image-only) cannot be parsed
• Password-protected PDFs are not supported
• Try converting to DOCX first

Privacy & Security

• No internet required after model download
• No telemetry or usage tracking
• No data collection - documents processed in memory only
• Open source - audit the code yourself

License

MIT License - See LICENSE file for details.

Acknowledgments

• llama.cpp - Efficient LLM inference
• Mistral AI - Base model
• TheBloke - Quantized models
• CustomTkinter - Modern GUI toolkit