Minima provides an on-premises, containerized Retrieval Augmented Generation (RAG) solution for querying local documents. It targets users who need to keep their data private while integrating with popular LLMs like ChatGPT or Anthropic Claude, or operate entirely offline.

How It Works

Minima utilizes a containerized architecture, allowing flexible deployment modes. It supports fully isolated local operation using Ollama for LLM inference, or integration with external services like ChatGPT or Anthropic Claude. The system indexes local documents (PDF, XLS, DOCX, TXT, MD, CSV) using Sentence Transformer embedding models and a specified reranker, storing embeddings in Qdrant.

Quick Start & Requirements

• Installation: Use docker compose with specific files (docker-compose-ollama.yml, docker-compose-chatgpt.yml, docker-compose-mcp.yml) and a .env file. For Claude Desktop integration, npx -y @smithery/cli install minima --client claude can be used.
• Prerequisites: Docker, Python >= 3.10 (for MCP), uv (for MCP). Requires specifying LOCAL_FILES_PATH, EMBEDDING_MODEL_ID, EMBEDDING_SIZE, OLLAMA_MODEL (for local), RERANKER_MODEL, USER_ID, and PASSWORD (for ChatGPT).
• Resources: Requires local compute for embedding, reranking, and potentially LLM inference.
• Docs: Minima GitHub

Highlighted Details

• Supports three distinct operational modes: fully local, ChatGPT integration, and Anthropic Claude integration.
• Configurable via environment variables for embedding models, LLMs, and rerankers.
• Provides a local chat UI accessible at http://localhost:3000 for the fully local setup.

Maintenance & Community

• Project maintained by dmayboroda.
• No explicit community links (Discord/Slack) or roadmap mentioned in the README.

Licensing & Compatibility

• Licensed under the Mozilla Public License v2.0 (MPLv2).
• MPLv2 is generally permissive for commercial use and linking with closed-source software, but requires modifications to the licensed code to be shared under the same license.

Limitations & Caveats

The project is described as "open source RAG on-premises containers," implying it may still be under active development. Specific LLM and reranker compatibility beyond tested models (e.g., sentence-transformers/all-mpnet-base-v2, BAAI rerankers) is not detailed.