Minima

Minima lets you ask questions about your own documents, with a choice of local inference or an external model. It's a self-hosted RAG tool for people and teams who want document-based answers while controlling where indexing and AI processing happen. It runs in Docker containers and is open source under the Mozilla Public License 2.0.

The fully local mode uses Ollama and keeps the language model, embeddings and reranker on your PC or your own server. It doesn't depend on ChatGPT or Claude. If you already host a model, Minima can use an OpenAI-compatible API through vLLM, TGI, Ollama Server or LocalAI. In that mode, indexing stays local, while inference runs wherever your chosen endpoint lives, including a cloud server. The OpenAI API is also supported.

Minima indexes folders and their subfolders, covering PDF, XLS, DOCX, plain text, Markdown and CSV files. It uses Sentence Transformer models for embeddings and Qdrant for document retrieval. The Ollama mode reranks search results before generating an answer; the custom model mode lets the model decide when to search documents and uses direct retrieval without reranking. A browser chat interface and an Electron app provide ways to ask questions.

For people who prefer an existing assistant, Minima connects document search to ChatGPT custom GPTs, Claude Desktop and GitHub Copilot through MCP. ChatGPT and Claude remain cloud services in those integrations; keeping the document index on your machine doesn't make their inference local.