Perceive Search 🚀

Privacy-respecting metasearch engine with integrated AI summaries powered by local open-source LLMs.

Perceive Search proxies Google/Bing results (ad/tracker-free) + generates concise AI insights for queries. Fully self-hosted, lightweight, and offline-capable after setup.

How It's Done

This project is built by forking the original Whoogle Search code and enhancing it with modern AI capabilities. I kept the privacy-focused core of Whoogle but added a lightweight open-source Large Language Model (LLM) Qwen2.5-0.5B-Instruct.

How It Works

1. Search Proxy: When you search, the backend (Python/Flask) fetches results from Google without passing your personal data (IP, cookies, etc.).
2. AI Summarization: The top search results are fed into the local Qwen2.5 model running on your CPU. The model reads the snippets and generates a concise answer to your query.
3. Modern UI: A custom React frontend displays the results and the AI summary in a clean, "Perceive" themed interface.

Installation

Follow these steps to set up Perceive Search on your own system.

Prerequisites

• Python
• Node.js & npm

1. Clone the Repository

First, download the code to your machine:

git clone git clone https://github.com/yourusername/perceive-search.git
cd perceive-search

2. Backend Setup

Set up the Python backend and download the AI model.

cd backend

# Install required Python libraries
pip install -r requirements.txt

# Download the AI model (Qwen2.5-0.5B-Instruct)
# This saves the model locally so it runs offline later. This is a lightweight model that runs on most CPUs as well.
python3 download_model.py

# Return to root
cd ..

3. Frontend Setup

Install the dependencies for the user interface.

cd frontend

# Install Node.js packages
npm install

# Return to root
cd ..

4. Run Perceive Search

I have included a simple script to start everything at once.

# Make the script executable (only needed once)
chmod +x start.sh

# Start the application
./start.sh

You will see output indicating that the backend (port 5001) and frontend (port 5173) are starting.

Open your browser to: http://localhost:5173

Features

• Privacy First: No ads, no tracking, no cookies, no IP logging.
• AI Powered: Instant summaries for your queries using a local LLM.
• Lightweight: Runs efficiently on standard CPUs (no GPU required).
• Modern Design: A beautiful, dark-themed interface with 3D elements and smooth animations.
• Self-Hosted: You own the data and the infrastructure.

Credits & License

This project is inspired by and enhanced from Whoogle Search by Ben Busby.

• Original Author: Ben Busby
• Original License: [MIT License] © 2020 Ben Busby
• My Enhancements: LLM integration (Qwen2.5-0.5B-Instruct) Frontend (React) Backend (Python/Flask)
• My License: [MIT License] © 2025 Vikram Lingam.