Minima is an open-source, on-premises RAG solution that runs in containers and can integrate with ChatGPT and MCP. It supports three modes: an isolated on-prem deployment with all components (LLM, embedding, and reranker) running locally; Custom GPT mode, where you query your local documents through the ChatGPT app or web with custom GPTs while the indexer operates on your PC or cloud and ChatGPT acts as the primary LLM; and Anthropic Claude mode, which uses Claude as the primary LLM with a local indexer. It indexes documents from LOCAL_FILES_PATH recursively and supports file types such as .pdf, .xls, .docx, .txt, .md, and .csv. The stack is configurable via EMBEDDING_MODEL_ID and EMBEDDING_SIZE for Qdrant storage, OLLAMA_MODEL for the LLM, and RERANKER_MODEL for reranking, with USER_ID for authentication and PASSWORD for a Firebase account. Minima offers multiple docker-compose options for fully local, ChatGPT-enabled, and MCP-integrated deployments, plus guidance for MCP usage with Anthropic Desktop and Copilot. An MCP workflow can also be installed via Smithery.