docker-compose.yml

# RagIndex — Turing Tree — full local stack in three services:
#   ollama   : the local LLM runtime (serves the qwen2.5 chat + qwen3 embeddings)
#   backend  : the FastAPI app over the sockets (vectorless RAG, branch search)
#   frontend : the Vite-built UI served by nginx (also proxies /api -> backend)
#
# ---------------------------------------------------------------------------
# Quick start (CPU — works everywhere):
#   docker compose up -d --build
#   # the one-shot 'ollama-pull' service downloads the models automatically,
#   # then open http://localhost:5173
#
# GPU (NVIDIA — optional; needs the NVIDIA Container Toolkit on the host):
#   docker compose -f docker-compose.yml -f docker-compose.gpu.yml up -d --build
#
# Everything runs locally — no API keys. Models live in the named 'ollama'
# volume, so they are pulled only once. Override models/ports with a .env file
# (see .env.example).
# ---------------------------------------------------------------------------
services:
  ollama:
    image: ollama/ollama:latest
    ports:
      - "${OLLAMA_PORT:-11434}:11434"
    volumes:
      - ollama:/root/.ollama
    healthcheck:
      test: ["CMD-SHELL", "ollama list >/dev/null 2>&1 || exit 1"]
      interval: 10s
      timeout: 5s
      retries: 15
      start_period: 20s
    restart: unless-stopped

  # One-shot helper: waits for Ollama, pulls the chat + embedding models into the
  # shared volume, then exits. Makes `up` turnkey — no manual `ollama pull`.
  ollama-pull:
    image: ollama/ollama:latest
    depends_on:
      ollama:
        condition: service_healthy
    environment:
      OLLAMA_HOST: http://ollama:11434
      RAGINDEX_CHAT_TAG: ${RAGINDEX_CHAT_TAG:-qwen2.5:3b-instruct}
      RAGINDEX_EMBED_TAG: ${RAGINDEX_EMBED_TAG:-qwen3-embedding:0.6b}
    entrypoint: ["/bin/sh", "-c"]
    command:
      - |
        set -e
        echo "Pulling chat model: $${RAGINDEX_CHAT_TAG}"
        ollama pull "$${RAGINDEX_CHAT_TAG}"
        echo "Pulling embedding model: $${RAGINDEX_EMBED_TAG}"
        ollama pull "$${RAGINDEX_EMBED_TAG}"
        echo "All models ready."
    restart: "no"

  backend:
    build:
      context: .
      dockerfile: backend.Dockerfile
    environment:
      OLLAMA_HOST: http://ollama:11434
      RAGINDEX_CHAT_TAG: ${RAGINDEX_CHAT_TAG:-qwen2.5:3b-instruct}
      RAGINDEX_EMBED_TAG: ${RAGINDEX_EMBED_TAG:-qwen3-embedding:0.6b}
      RAGINDEX_INDEX_TAG: ${RAGINDEX_INDEX_TAG:-qwen2.5:3b-instruct}
    depends_on:
      ollama:
        condition: service_healthy
      ollama-pull:
        condition: service_completed_successfully
    ports:
      - "${BACKEND_PORT:-8000}:8000"
    volumes:
      - ragdata:/app/data
    healthcheck:
      test: ["CMD-SHELL", "python -c \"import urllib.request; urllib.request.urlopen('http://localhost:8000/api/health', timeout=5)\" || exit 1"]
      interval: 15s
      timeout: 8s
      retries: 5
      start_period: 40s
    restart: unless-stopped

  frontend:
    build:
      context: ./frontend
    depends_on:
      backend:
        condition: service_healthy
    ports:
      - "${FRONTEND_PORT:-5173}:80"
    restart: unless-stopped

volumes:
  ollama:
  ragdata: