# ollama — local LLM/embedding runtime for the homelab.
#
# CPU-only deployment on herbydev. No GPU runtime (no `runtime: nvidia`,
# no `devices:` block). Reachable as `ollama:11434` from any container
# on the shared `homelab` network. Loopback-only on the host (no public
# exposure, no cloudflared route, no Tailscale binding).
#
# Models persist in the named `ollama-models` volume across restarts.
#
# Usage:
#   docker compose -f compose.yml up -d
#   docker exec ollama ollama pull <model>
#   docker exec ollama ollama list

name: ollama

services:
  ollama:
    image: ollama/ollama:latest
    container_name: ollama
    restart: unless-stopped
    environment:
      - OLLAMA_KEEP_ALIVE=${OLLAMA_KEEP_ALIVE}
      - OLLAMA_NUM_PARALLEL=${OLLAMA_NUM_PARALLEL}
    ports:
      - "127.0.0.1:11434:11434"  # loopback on host; in-network via Docker DNS
    volumes:
      - ollama-models:/root/.ollama
    networks:
      - homelab
    # No runtime/devices/gpus — CPU only by design.

volumes:
  ollama-models:

networks:
  homelab:
    external: true