PrivateGPT neural network

PrivateGPT is a production-ready artificial intelligence project that allows users to interact with their documents using the power of Large Language Models (LLM), even in situations where there is no Internet connection. This AI tool provides complete privacy as no data leaves the user's runtime during the interaction. The project provides an API that offers all the necessary primitives for building proprietary, context-aware AI applications. It follows and extends the OpenAI API standard, supporting both regular and stream responses.

The API is divided into two logical parts: a High-level API , which abstracts the complexity of implementing a Retrieval Augmented Generation (RAG) pipeline, and a Low-level API , which allows advanced users to implement their own complex pipelines. The high-level API includes functionality such as document loading, chat, and completion using context from loaded documents, and also abstracts context retrieval, query creation, and response generation. The low-level API provides functionality such as generating attachments and retrieving contextual snippets.

PrivateGPT is designed to address the significant data privacy challenge of using generative artificial intelligence by companies of all sizes and in data-sensitive fields such as healthcare or legal. It offers a solution to give an organization complete control over its data, eliminating the risk of data leakage associated with third-party AI tools.

The project architecture is built using FastAPI and follows the OpenAI API blueprint. It uses LlamaIndex abstractions such as LLM , BaseEmbedding and VectorStore , which makes it easy to change the actual implementation of these abstractions. PrivateGPT's design allows both the API and RAG implementation to be easily extended and adapted, providing simplicity and flexibility.

PrivateGPT is actively supported by the Qdrant, Fern and LlamaIndex teams, and has also received influence and support from other significant projects such as LangChain, GPT4All, LlamaCpp, Chroma and SentenceTransformers.