🔒 PrivateSearch: Wiki Edition

PrivateSearch is a semantic document search system, 100% local and private, inspired by Andrej Karpathy's "LLM Wiki" paradigm.

Unlike traditional RAG systems, the "Wiki Edition" distills the essence of every document into an encrypted summary, creating a high-density knowledge base protected by military-grade encryption.

🚀 Key Features

• Encrypted Vault (AES-256): All AI-generated data is encrypted using a Master Password. Without it, your wiki index remains unreadable.
• Essential Synthesis: The AI (via Ollama) reads your files (PDFs, Images via OCR, Text) and creates an ultra-concise "identity card" (max 20 words) focused on facts.
• Point & Open: Find a document via semantic chat and open it instantly using your PC's default application (Acrobat, Word, etc.) with a single click.
• Total Privacy: No data ever leaves your PC. Everything runs locally via the Ollama framework.
• Multi-Vault: Manage different document folders as separate and independent encrypted vaults.

🛠️ Requirements

• Linux (tested on Ubuntu)
• Ollama installed and running (ollama serve)
• Python 3.10+
• At least one Qwen2.5 or Llama3 model downloaded in Ollama (e.g., ollama pull qwen2.5:7b)

📦 Quick Start

1. Clone the repository to your local machine.
2. Ensure Ollama is running.
3. Run the startup script:

bash start.sh

The script will automatically create the virtual environment (.venv), install dependencies, and launch the web interface.

📖 How to Use

1. Unlock: Go to the "Access & Config" tab and set your Master Password.
2. Compile: In the "Wiki Management" tab, enter a local folder path and click "Compile / Update Wiki". The AI will start reading and synthesizing your files.
3. Search: Go to the "Wiki Search" tab. You'll see the complete file index on the right. On the left, you can ask questions like: "Find contracts expired in 2023" or "Is there a file mentioning Enel?".
4. Open: Click the "Open [filename]" buttons that appear below the chat to consult the original document.

🔐 Security

The system implements:

• PBKDF2 for key derivation from your password.
• AES-256 (Fernet) for .wiki.enc file encryption.
• Local OCR: Text recognition for images and scanned PDFs happens entirely on your machine.

---

PrivateSearch — Your knowledge, protected and searchable.