# Local LLM runner (Docker Model Runner-style).
# Inspired by GitLab Runner K8s agent layout: single service, config, volumes.
# Default: Ollama (OpenAI-compatible, run LLMs locally).
# Override image/command via env: LLM_IMAGE, LLM_COMMAND (see .env.llm.example).
# Ref: https://docs.ollama.com/docker
# Ref: https://www.docker.com/blog/run-llms-locally/
services:
llm-runner:
image: ${LLM_IMAGE:-ollama/ollama}
container_name: llm-runner
command: ${LLM_COMMAND:-} # optional; omit to use image default (Ollama serve)
environment:
- OLLAMA_HOST=0.0.0.0
- OLLAMA_ORIGINS=*
volumes:
- llm-data:/root/.ollama
ports:
- "${LLM_PORT:-11434}:11434"
restart: unless-stopped
networks:
- llm-network
# For NVIDIA GPU (optional): uncomment and ensure nvidia container runtime.
# deploy:
# resources:
# reservations:
# devices:
# - driver: nvidia
# count: all
# capabilities: [gpu]
networks:
llm-network:
driver: bridge
volumes:
llm-data:
name: llm-data
