Chat with your documents privately, locally, and securely ✨
Features • Installation • Usage • Architecture • Contributing
PrivateGPT transforms your personal documents into an intelligent, conversational AI assistant. Unlike cloud-based solutions, everything runs locally on your machine, ensuring your sensitive data never leaves your control.
- 🔐 Complete Privacy: Your documents never leave your machine
- 📚 Smart Document Understanding: Chat with PDFs, Word docs, and text files
- ⚡ Real-time Responses: Streaming responses for better user experience
- 🎯 Context-Aware: Provides accurate answers based on your documents
- 🛠️ Easy Setup: Simple installation and intuitive interface
- Supports multiple formats: PDF, DOCX, DOC, TXT
- Automatic text chunking for optimal retrieval
- Vector embeddings for semantic search
- Streamlit-based chat interface with design first philosophy
- Real-time streaming responses
- Context-aware conversations
- Emoji support for better readability
- RAG (Retrieval-Augmented Generation) pipeline
- ChromaDB for vector storage
- Ollama for local LLM inference
- Modular and extensible design
Before getting started, ensure you have the following installed:
- Python 3.8+ 🐍
- Ollama (for local LLM inference) 🤖
- Homebrew (for macOS users) 🍺
# macOS
brew install ollama
# Linux
curl -fsSL https://ollama.ai/install.sh | sh
# Windows (using WSL)
curl -fsSL https://ollama.ai/install.sh | sh# Download the embedding model
ollama pull nomic-embed-text
# Download the language model
ollama pull mistral# Clone the repository
git clone https://github.com/dangkv/privateGPT.git
cd privateGPT
# Create virtual environment
python -m venv venv
source venv/bin/activate # On Windows: venv\Scripts\activate
# Install dependencies
pip install -r requirements.txtstreamlit>=1.28.0
langchain>=0.0.350
langchain-community>=0.0.10
chromadb>=0.4.15
ollama>=0.1.0
pypdf>=3.17.0
python-docx>=0.8.11
docx2txt>=0.8Understanding the data folder structure is crucial for setting up PrivateGPT:
data/
├── raw/ # 📄 Your unstructured documents go here
│ ├── document1.pdf
│ ├── report.docx
│ ├── notes.txt
│ └── subfolder/
│ └── more_docs.pdf
├── processed/ # 🔧 Processed document chunks (auto-generated)
└── chroma_db/ # 🗄️ Vector database storage (auto-generated)
├── chroma.sqlite3
└── index/
-
Create the data directory (if it doesn't exist):
mkdir -p data/raw
-
Add your documents to the
data/raw/folder:- Supported formats:
.pdf,.docx,.doc,.txt - You can organize documents in subfolders
- No size limit (within reason)
- Supported formats:
-
Example structure:
data/raw/ ├── company_policies/ │ ├── employee_handbook.pdf │ └── code_of_conduct.docx ├── research_papers/ │ ├── ai_trends_2024.pdf │ └── machine_learning_guide.pdf └── personal_notes/ ├── meeting_notes.txt └── project_ideas.docx
-
Start the services:
make start
-
Ingest your documents:
make db-ingest
-
Open your browser and go to
http://localhost:8501 -
Start chatting with your documents! 💬
-
Shut down bot:
make stop
PrivateGPT comes with a comprehensive Makefile for easy management:
# 🎯 Main Commands
make start # Start all services
make stop # Stop all services
make status # Check service status
# 📊 Streamlit Management
make streamlit-start # Start Streamlit app
make streamlit-stop # Stop Streamlit app
make streamlit-reset # Restart Streamlit
# 🤖 Ollama Management
make llm-start # Start Ollama service
make llm-stop # Stop Ollama service
make llm-status # Check Ollama status
# 🗄️ Database Management
make db-ingest # Process and ingest documents
make db-reset # Reset database
make db-backup # Create database backup
make db-check # Check database status
# 🧹 Cleanup
make clean # Clean temporary files
make clean-all # Deep clean everythingIf you prefer to run the ingestion manually:
python scripts/ingest_documents.pyPrivateGPT follows a modular RAG (Retrieval-Augmented Generation) architecture:
┌─────────────────┐ ┌─────────────────┐ ┌─────────────────┐
│ 📄 Documents │ ── │ 📝 Ingestion │ ── │ 🗄️ Vector DB │
│ (PDF, DOCX, │ │ (Chunking & │ │ (ChromaDB) │
│ TXT files) │ │ Embedding) │ │ │
└─────────────────┘ └─────────────────┘ └─────────────────┘
│
┌─────────────────┐ ┌─────────────────┐ ┌─────────────────┐
│ 💬 Streamlit │ ── │ 🧠 RAG │ ── │ 🔍 Retrieval │
│ Interface │ │ Pipeline │ │ (Semantic │
│ │ │ │ │ Search) │
└─────────────────┘ └─────────────────┘ └─────────────────┘
│
┌─────────────────┐
│ 🤖 Generation │
│ (Ollama + │
│ Mistral) │
└─────────────────┘
- 🔄 Ingestion Pipeline: Processes documents and creates vector embeddings
- 🔍 Retrieval System: Finds relevant document chunks using semantic search
- 🤖 Generation Engine: Uses Ollama with Mistral model for response generation
- 💬 Chat Interface: Streamlit-based UI with real-time streaming
- 🗄️ Vector Database: ChromaDB for efficient similarity search
Customize PrivateGPT by modifying config.py:
# Model configurations
EMBEDDING_MODEL = "nomic-embed-text" # Change embedding model
LLM_MODEL = "mistral" # Change language model
OLLAMA_BASE_URL = "http://localhost:11434"
# Chunking parameters
CHUNK_SIZE = 1000 # Adjust chunk size
CHUNK_OVERLAP = 200 # Adjust overlap
# Retrieval parameters
TOP_K_RETRIEVAL = 5 # Number of documents to retrieve- Follow PEP 8 style guidelines
- Add docstrings to all functions and classes
- Include type hints where appropriate
- Write unit tests for new features
- Update documentation as needed
This project is licensed under the MIT License - see the LICENSE file for details.
- LangChain for the RAG framework
- Ollama for local LLM inference
- ChromaDB for vector storage
- Streamlit for the web interface
- Mistral for the language model