docker-compose.yml

services:
  ollama-pull:
    image: alpine/ollama
    container_name: ollama-pull
    environment:
      OLLAMA_HOST: localhost:11434  # instead of IP 127.0.0.1
    env_file:
      - env.local
    entrypoint: sh
    command: -c 'env | grep _MODEL | cut -d= -f2 | sed "s/^[^/]*\///" | xargs -I{} ollama pull {}'
    extra_hosts:  # send localhost traffic to the docker host, e.g. your laptop
      - "localhost:host-gateway"

  llama-stack:
    depends_on:
      ollama-pull:
        condition: service_completed_successfully
    image: llamastack/distribution-starter:0.5.0
    container_name: llama-stack
    tty: true
    env_file:
      - env.local
    ports:
      - "8321:8321"
    # Ensure the container which specially treats localhost routes back to the
    # host machine, e.g. your laptop.
    environment:
      - OPENAI_BASE_URL=http://host.docker.internal:11434/v1
      - OTEL_EXPORTER_OTLP_ENDPOINT=http://host.docker.internal:4318
      # Ensure we only see traces sampled upstream. This reduces noise without
      # disabling SQL tracing entirely.
      - OTEL_TRACES_SAMPLER=parentbased_traceidratio
      - OTEL_TRACES_SAMPLER_ARG=0.0
    extra_hosts:
      - "host.docker.internal:host-gateway"