# Fully self-contained, reproducible run: Ollama + agent in one command.
#   docker compose up --build
#
# NOTE: Inside Docker, Ollama runs CPU-only (Docker Desktop on macOS cannot
# reach the Metal GPU), so first run also downloads ~6.6 GB and inference is
# slow. For a fast run, instead run Ollama natively on the host and use the
# `docker run ... host.docker.internal` path documented in the README.
services:
  ollama:
    image: ollama/ollama:latest
    volumes:
      - ollama-models:/root/.ollama
    ports:
      - "11434:11434"

  agent:
    build: .
    depends_on:
      - ollama
    environment:
      AGENT_OLLAMA_HOST: http://ollama:11434
      AGENT_OLLAMA_MODEL: qwen3.5:9b
      AGENT_AUTO_PULL: "1" # pull the model on first run if missing

volumes:
  ollama-models: