dmayboroda/minima

./run.sh (interactive setup with mode selection)
docker compose -f docker-compose-ollama.yml --env-file .env up --build (fully local)
docker compose -f docker-compose-custom-llm.yml --env-file .env up --build (custom LLM)
docker compose -f docker-compose-chatgpt.yml --env-file .env up --build (ChatGPT)
docker compose -f docker-compose-mcp.yml --env-file .env up --build (Anthropic Claude MCP)

Usage

• •Create .env from .env.sample with LOCAL_FILES_PATH, EMBEDDING_MODEL_ID, EMBEDDING_SIZE; add OLLAMA_MODEL + RERANKER_MODEL for Ollama mode, or LLM_BASE_URL + LLM_MODEL for custom LLM mode
• •For Anthropic Claude: add the minima MCP server to ~/Library/Application Support/Claude/claude_desktop_config.json using uv --directory /path_to_cloned_minima_project/mcp-server run minima
• •For ChatGPT: copy the OTP from the docker terminal and use the 'Minima Local Computer Search' custom GPT
• •Chat UI is served at http://localhost:3000; fully local users can also run the Electron app (cd electron && npm install && npm start)
• •Indexed file types: .pdf, .xls, .docx, .txt, .md, .csv (recursive over LOCAL_FILES_PATH)

Key features

• ✓Four modes: fully local Ollama (all NNs on your hardware), custom OpenAI-compatible LLM (vLLM, TGI, Ollama server, LiteLLM, LocalAI, OpenAI API), ChatGPT custom GPT, and Anthropic Claude
• ✓Ollama workflow: query enhancement → retrieval with BAAI reranker reranking → answer generation
• ✓Custom LLM workflow: LLM decides via function calling whether to search → direct vector search (no reranking) → answer
• ✓Qdrant vector storage configured from EMBEDDING_SIZE; Sentence Transformer embedding models
• ✓MCP usage with GitHub Copilot via run_in_copilot.sh and .vscode/mcp.json

Caveats

• ⚠Only Sentence Transformer embedding models are supported (tested with sentence-transformers/all-mpnet-base-v2)
• ⚠Reranker model is NOT required and is not downloaded in custom LLM mode
• ⚠For MCP usage your local machine needs Python >= 3.10 and 'uv' installed
• ⚠ChatGPT integration requires USER_ID (your email) and PASSWORD env vars
• ⚠License: Mozilla Public License v2.0 (MPLv2)