RAG-GPT provides a comprehensive solution for building intelligent customer service systems using Retrieval-Augmented Generation (RAG) and Large Language Models (LLMs). It targets developers and businesses seeking to quickly deploy a customizable chatbot with a user-friendly interface and the ability to learn from diverse knowledge bases.

How It Works

The system leverages a Flask backend to integrate various LLMs (OpenAI, ZhipuAI, DeepSeek, Moonshot, Ollama) and supports multiple knowledge base types, including websites, isolated URLs, and local files. It employs a RAG architecture for contextually relevant answers, with options for query preprocessing and result reranking to enhance accuracy.

Quick Start & Requirements

• Install: git clone the repository, configure .env with API keys and model names, then run docker-compose up --build or python3 rag_gpt_app.py (after pip install -r requirements.txt and python3 create_sqlite_db.py).
• Prerequisites: Python 3.10+, API keys for chosen LLMs, optionally Ollama for local LLMs.
• Setup Time: Claimed to be as fast as five minutes for a production-level service.
• Links: Live Demo, Admin Console

Highlighted Details

• Supports cloud-based and local LLMs.
• Integrates websites, isolated URLs, and local files (txt, md, pdf, docx, etc., up to 30MB per file).
• Offers an attractive, customizable UI for both the chatbot and an admin console.
• Includes a dashboard for viewing user historical requests.
• Provides an iframe embed option for website integration.

Maintenance & Community

The project is hosted on GitHub under the open-kf organization. Specific contributor or community links (Discord, Slack, roadmap) are not detailed in the README.

Licensing & Compatibility

The README does not explicitly state a license. Compatibility for commercial use or closed-source linking is not specified.

Limitations & Caveats

The README mentions that only gpt-3.5-turbo is currently available in the admin console's LLM selection, with plans for expansion. DeepSeek and Moonshot require ZhipuAI's Embedding API, as they do not provide their own.