searchGPT is an open-source RAG-based search engine designed to provide grounded, natural language answers to user queries. It targets users seeking factual information by integrating LLMs with real-time web search and local file content analysis, offering a more reliable alternative to ungrounded LLM responses.

How It Works

This project implements a Retrieval-Augmented Generation (RAG) architecture. It retrieves relevant information from web search results or local files (PPT, DOC, PDF) and then uses this context to inform LLM responses. This approach addresses the LLM's knowledge limitations by providing real-time, factual data, thereby reducing hallucinations and improving answer accuracy, as demonstrated by its comparison to ungrounded answers.

Quick Start & Requirements

• Installation: pip install -r requirements.txt (within a Python 3.10.8 environment).
• Prerequisites: OpenAI API Key or GooseAI API Key, Azure Bing Search Subscription Key.
• Setup: Configure API keys in backend/src/config/config.yaml. Run app.py for the web UI or main.py for command-line output.
• Demo: https://searchgpt-demo.herokuapp.com/index

Highlighted Details

• Supports web search with real-time results and file content search (PPT/DOC/PDF).
• Integrates semantic search capabilities using FAISS or PyTerrier.
• Compatible with OpenAI and GooseAI LLM providers.
• Features an intuitive, easy-to-use frontend UI.

Maintenance & Community

The project welcomes contributions, particularly from frontend developers. Further details are available in their contributing guidelines.

Licensing & Compatibility

Licensed under the MIT License, permitting commercial use and integration with closed-source projects.

Limitations & Caveats

The demo page has a loading time of approximately 10 seconds and requests users not to abuse it with automated programs.