zylon-ai/private-gpt

PrivateGPT is a production-ready Python application that lets you ask questions about your own documents using large language models (LLMs) while keeping all of your data completely private. The core problem it solves is this: tools like ChatGPT are powerful, but they require sending your data to third-party servers, a serious concern for healthcare providers, law firms, banks, and other organizations handling sensitive information. PrivateGPT runs entirely on your own machine or server, so no data ever leaves your environment. Under the hood, PrivateGPT uses a technique called Retrieval Augmented Generation, or RAG. When you upload documents, the system parses and splits them into chunks, generates numerical representations called embeddings, and stores everything locally. When you ask a question, it retrieves the most relevant chunks and feeds them to the LLM alongside your question, producing an answer grounded in your actual documents rather than the model's training data alone. The project exposes two API layers. The high-level API handles document ingestion and chat with minimal setup. The low-level API gives developers direct access to embeddings and chunk retrieval so they can build custom workflows on top of the same infrastructure. A ready-to-use chat interface built with Gradio is also included for testing without writing any code. You would reach for PrivateGPT when you need to search or interrogate internal documents, contracts, reports, manuals, research files, and cannot or will not use a cloud AI service. It works offline, making it suitable for air-gapped environments. Technically, the backend is a FastAPI server (Python), the RAG pipeline is powered by LlamaIndex, and it follows the OpenAI API standard so it integrates with any client that already speaks that protocol.