RAGstack provides infrastructure to deploy a private ChatGPT alternative within a VPC. It connects LLMs (GPT4All for local CPU deployment, Falcon-7b and Llama 2 for cloud GPU clusters) to organizational knowledge bases via retrieval augmented generation. The system retrieves relevant information from external sources like SaaS apps and documents, injecting it into the LLM context window to extend model knowledge beyond training data.