GitHub - kbrisso/byte-vision: Byte-Vision is a privacy-first document intelligence platform that transforms static documents into an interactive, searchable knowledge base. Built on Elasticsearch with RAG (Retrieval-Augmented Generation) capabilities, it offers document parsing, OCR processing, and a modern UI.

9 min read Original article ↗

Byte-Vision

Byte-Vision: AI-Powered Document Intelligence Platform

Status: Beta - Under active development

Byte-Vision is a privacy-first document intelligence platform that transforms static documents into an interactive, searchable knowledge base. Built on Elasticsearch with RAG (Retrieval-Augmented Generation) capabilities, it offers document parsing, OCR processing, and conversational AI interfaces—all running locally to ensure complete data privacy.

✨ Key Features

  • 📄 Universal Document Processing - Parse PDFs, text files, and CSVs with built-in OCR for image-based content
  • 🔍 AI-Enhanced Search - Semantic search powered by Elasticsearch and vector embeddings
  • 💬 Conversational AI - Document-specific Q&A and free-form chat with local LLM integration
  • 📊 Research Management - Automatically save and organize insights from document analysis
  • 🔒 Privacy-First - Runs entirely locally with no external data transmission
  • 🖥️ Intuitive Interface - Full-featured UI that simplifies complex document operations

🚀 Quick Start

Prerequisites

Installation

For detailed setup instructions, see Installation Guide.

📋 Table of Contents

🖼️ Interface Tour

Document Search Screen

The main "Document Search" screen allows you to locate and analyze documents after they have been parsed and indexed in Elasticsearch.

"Document Search"

Document Viewer

Click the "View" button to display the original parsed document.

"PDF View"

Question and Answer Interface

Default View - History Tab

View previously saved question-answer history items for the selected document.

"Document Ask"

Question Entry Form

Enter your questions about the document using this interface.

doc-qa.png

Processing Stage

The system processes your question and searches through the document.

doc-qa-2.png

Results Display

View the AI-generated answers based on your document content.

doc-qa-3.png

Export to PDF

Export your question-answer sessions to PDF format for documentation.

"QA PDF Export"

Document Processing Features

Document Parsing and Chunking

Parse PDF, text, and CSV files for processing and analysis.

process-ocr-text.png

Parser Results

View the results of document parsing and chunking operations.

doc-parse-results.png

OCR Processing

Image Scan Setup

Configure OCR settings for processing scanned documents.

ocr-processing-1.png

OCR Results

Review extracted text from image-based documents.

ocr-processing-2.png

AI Inference Screen

Main Interface

Primary inference screen for general AI conversations.

inference-prompt-1.png

Chat History

View previous conversations and responses.

inference-prompt-2.png

Export Chat History

Export inference conversations to PDF format.

inference-prompt-pdf.png

inference-prompt-pdf-2.png

📦 Installation

Prerequisites

Component Version Purpose
Go 1.23+ Backend services
Node.js 18+ Frontend build system
Elasticsearch 8.x Document indexing and search
Wails v2 Desktop application framework

System Requirements

  • OS: Windows 10+, macOS 10.13+, or Linux
  • RAM: 8GB minimum (16GB recommended)
  • Storage: 5GB free space
  • CPU: Multi-core processor recommended

Optional Dependencies

  • CUDA: Enables GPU acceleration for AI models
  • Docker: Containerize Elasticsearch for easier deployment

Development Setup

1. Clone and Install Dependencies

git clone https://github.com/kbrisso/byte-vision.git
cd byte-vision

# Install Go dependencies
go mod download && go mod tidy

# Install Wails CLI
go install github.com/wailsapp/wails/v2/cmd/wails@latest

# Install frontend dependencies
cd frontend && npm install && cd ..

2. Set Up Elasticsearch

Option A: Docker (Recommended)

-p 9200:9200 -p 9300:9300
-e "discovery.type=single-node"
-e "xpack.security.enabled=false"
docker.elastic.co/elastic/elastic:8.11.0

Option B: Local Installation

  1. Download from Elasticsearch Downloads
  2. Extract and run:
# Windows
bin\elasticsearch.bat       
# macOS/Linux
bin/elastic

3. Install LlamaCpp

Option A: Download Pre-built Binaries (Recommended)

  1. Visit LlamaCpp releases
  2. Download for your platform:
    • Windows: llama-*-bin-win-x64.zip (CPU) or llama-*-bin-win-cuda-cu*.zip (GPU)
    • Linux: llama-*-bin-ubuntu-x64.tar.gz
    • macOS: brew install llama.cpp
  3. Extract to llamacpp/ directory

Option B: Build from Source

git clone https://github.com/ggerganov/llama.cpp.git temp-llama
cd temp-llama && mkdir build && cd build
cmake .. -DLLAMA_CUDA=ON  # Add for GPU support
cmake --build . --config Release
cp bin/llama-cli ../llamacpp/
cd ../.. && rm -rf temp-llama

4. Download AI Models

Download example models

curl -L -o models/llama-2-7b-chat.Q4_K_M.gguf \
https://huggingface.co/TheBloke/Llama-2-7B-Chat-GGUF/resolve/main/llama-2-7b-chat.Q4_K_M.gguf

curl -L -o models/all-MiniLM-L6-v2.gguf \
https://huggingface.co/sentence-transformers/all-MiniLM-L6-v2-gguf/resolve/main/all-MiniLM-L6-v2.gguf

5. Install xpdf-tools

Download and install xpdf-tools for PDF processing:

Option A: Download Pre-built Binaries (Recommended)

  1. Visit Xpdf downloads
  2. Download the appropriate version for your platform:
    • Windows: xpdf-tools-win-*-setup.exe
    • Linux: xpdf-tools-linux-*-static.tar.gz
    • macOS: xpdf-tools-mac-*-setup.dmg
  3. Extract or install to the xpdf-tools/ directory in your project root

Option B: Package Manager Installation

# macOS
brew install xpdf
# Ubuntu/Debian
sudo apt-get install xpdf-utils
# Windows (using Chocolatey)
choco install xpdf-utils

6. Install Tesseract-OCR

Install Tesseract-OCR for optical character recognition:

Windows:

  1. Download from Tesseract releases
  2. Install the executable
  3. Add Tesseract to your system PATH:
    • Add C:\Program Files\Tesseract-OCR to your PATH environment variable
    • Or add custom path in byte-vision-cfg.env: TESSERACT_PATH=C:\path\to\tesseract.exe

macOS:

Linux (Ubuntu/Debian):

sudo apt-get install tesseract-ocr

Verify Installation:

7. Configure Environment

Create byte-vision-cfg.env:

Elasticsearch Configuration

ELASTICSEARCH_URL=http://localhost:9200 ELASTICSEARCH_USERNAME=elastic ELASTICSEARCH_PASSWORD=your_password

LlamaCpp Configuration

LLAMA_CLI_PATH=./llamacpp/llama-cli LLAMA_EMBEDDING_PATH=./llamacpp/llama-embedding

Model Configuration

MODEL_PATH=./models DEFAULT_INFERENCE_MODEL=llama-2-7b-chat.Q4_K_M.gguf DEFAULT_EMBEDDING_MODEL=all-MiniLM-L6-v2.gguf

Application Settings

MAX_CHUNK_SIZE=1000 CHUNK_OVERLAP=200 LOG_LEVEL=INFO

8. Run the Application

The application will launch with hot reload enabled.

Production Build

The built application will be in the build/ directory

⚙️ Configuration

The application uses environment variables defined in byte-vision-cfg.env:

Variable Description Default
ELASTICSEARCH_URL Elasticsearch server URL http://localhost:9200
ELASTICSEARCH_USERNAME Elasticsearch username elastic
ELASTICSEARCH_PASSWORD Elasticsearch password -
LLAMA_CLI_PATH Path to llama-cli executable ./llamacpp/llama-cli
LLAMA_EMBEDDING_PATH Path to llama-embedding executable ./llamacpp/llama-embedding
MODEL_PATH Directory containing AI models ./models
DEFAULT_INFERENCE_MODEL Default model for inference -
DEFAULT_EMBEDDING_MODEL Default model for embeddings -
MAX_CHUNK_SIZE Maximum text chunk size 1000
CHUNK_OVERLAP Overlap between chunks 200
LOG_LEVEL Application log level INFO

📝 Usage

First-Time Setup

  1. Start Elasticsearch: Ensure Elasticsearch is running
  2. Launch Byte-Vision: Run the application
  3. Configure Models: Go to Settings → LlamaCpp Settings and set paths
  4. Test Connection: Verify Elasticsearch connection in Settings

Document Management

  1. Upload Documents: Use the document parser to upload and process files
  2. Configure Chunking: Adjust text chunking settings for optimal search
  3. Index Documents: Process documents for embedding and search

Document-Specific Q&A

  • Select a document from the search results
  • Click "Ask Questions" to open the Q&A interface
  • Enter your questions and receive AI-generated answers
  • View answer sources and confidence scores
  • Export Q&A sessions to PDF

AI Interactions

  1. Ask Questions: Use the document question modal to query your documents
  2. Export Results: Export chat history to PDF for documentation
  3. Compare Responses: Use the comparison feature to evaluate different model outputs

Free-Form Chat

  • Access the AI Inference screen for general conversations
  • Chat with your local LLM models
  • Export conversation history
  • Compare different model responses

🔧 Troubleshooting

Common Issues

❌ Elasticsearch Connection Failed

Symptoms: Cannot connect to Elasticsearch service

Solutions:

  1. Verify Elasticsearch is running: 
    curl http://localhost:9200
  2. Check if port 9200 is available:
  3. Verify configuration in byte-vision-cfg.env
  4. Check firewall settings
  5. For Docker: Ensure container is running
❌ LlamaCpp Model Loading Error

Symptoms: Model fails to load or produces errors

Solutions:

  1. Verify model file exists in models/ directory
  2. Check model format (must be .gguf)
  3. Ensure sufficient RAM for model size
  4. Verify LLAMA_CLI_PATH in configuration
  5. Test LlamaCpp directly: 
    ./llamacpp/llama-cli --model ./models/your-model.gguf --prompt "Hello"
❌ Frontend Build Errors

Symptoms: npm install or build failures

Solutions:

  1. Clear npm cache: 
    cd frontend
rm -rf node_modules package-lock.json
npm cache clean --force
npm install
  2. Check Node.js version: node --version
  3. Update npm: npm install -g npm@latest
❌ Port Already in Use

Symptoms: Application fails to start due to port conflicts

Solutions:

  1. Find process using port: 
    # Windows
netstat -ano | findstr :3000

# macOS/Linux
lsof -ti:3000
  2. Kill process: 
    # Windows
taskkill /PID <PID> /F

# macOS/Linux
kill -9 <PID>

Performance Tips

  • GPU Acceleration: Install CUDA/ROCm for faster model inference
  • Model Selection: Use smaller quantized models for better performance
  • Memory Management: Adjust Elasticsearch heap size for large document collections
  • Chunking Optimization: Tune MAX_CHUNK_SIZE and CHUNK_OVERLAP for your use case

Debug Mode

Enable debug logging:

Check logs in ./logs/ directory for detailed error information.

🛠️ Development

Built With

Core Technologies

  • Wails - Desktop application framework
  • Go - Backend services and APIs
  • React - Frontend user interface
  • Elasticsearch - Document indexing and search
  • Llama.cpp - Local AI model inference

Frontend Stack

Backend Libraries

Project Structure

byte-vision/ 
├── 📁 build/ # Built application files 
├── 📁 document/ # Document storage 
├── 📁 frontend/ # React frontend source 
│ ├── 📁 src/ 
│ └── 📁 public/ 
├── 📁 llamacpp/ # LlamaCpp binaries 
├── 📁 logs/ # Application logs 
├── 📁 models/ # AI model files (.gguf) 
├── 📁 prompt-cache/ # Cached prompts 
├── 📁 prompt-temp/ # Prompt templates 
├── 📁 xpdf-tools/ # PDF processing tools 
├── 📄 byte-vision-cfg.env # Configuration file 
├── 📄 wails.json # Wails configuration 
└── 📄 go.mod # Go dependencies

Logs and Debugging

  • Application logs: ./logs/
  • Elasticsearch logs: Check Elasticsearch installation directory
  • Debug mode: wails dev -debug
  • Frontend logs: Browser developer console
  • Backend logs: Terminal output during development

🤝 Contributing

Contributions are what make the open source community such an amazing place to learn, inspire, and create. Any contributions you make are greatly appreciated.

If you have a suggestion that would make this better, please fork the repo and create a pull request. You can also open an issue with the tag "enhancement." Remember to give the project a star! Thanks again!

How to Contribute

  1. Fork the Project
  2. Create your Feature Branch (git checkout -b feature/AmazingFeature)
  3. Commit your Changes (git commit -m 'Add some AmazingFeature')
  4. Push to the Branch (git push origin feature/AmazingFeature)
  5. Open a Pull Request

Development Guidelines

  • Follow Go formatting standards (go fmt)
  • Write tests for new features
  • Update documentation for API changes
  • Use semantic commit messages
  • Ensure all tests pass before submitting

📈 Roadmap

In Progress

  • Settings persistence for llama-cli configuration
  • Settings persistence for llama-embedding configuration
  • Enhanced documentation and examples

Planned Features

  • Additional document format support (DOCX, PPT, etc.)
  • Advanced search filters and operators
  • Batch document processing capabilities
  • RESTful API for external integrations
  • Docker deployment configuration
  • User authentication and access control
  • Cloud storage integration (S3, Google Drive, etc.)
  • Multi-language support
  • Advanced analytics and reporting

Long-term Vision

  • Distributed processing for large document collections
  • Plugin architecture for custom processors
  • Integration with external AI services
  • Mobile application companion

See open issues for detailed feature requests and bug reports.

📄 License

This project is licensed under the terms of the MIT license.

📧 Contact

Kevin Brisson - LinkedIn - kbrisso@gmail.com

Project Link: https://github.com/kbrisso/byte-vision

⭐ Star this project if you find it helpful!