docker-compose.yml

services:
  web:
    build:
      context: .
      dockerfile: Dockerfile_cpu
    # CPU image: runs on linux/amd64 and linux/arm64 (incl. Docker Desktop on
    # Apple Silicon). For NVIDIA GPUs use docker-compose-cuda.yml instead.
    image: ghcr.io/katherlab/llmaix-cpu:latest
    ports:
      - "19999:${PORT:-5000}"
    volumes:
      # Default: named volume that persists Hugging Face model downloads
      # (the bundled config.yml pulls its model from Hugging Face automatically).
      # To use your own local model files + config instead, replace with a bind
      # mount (- /path/to/your/local/model/dir:/models) and set
      # CONFIG_FILE=/models/config.yml below.
      - llmaix_models:/models
    environment:
      # Path to the llama-server binary inside the official image. Don't change.
      - SERVER_PATH=${SERVER_PATH:-/app/llama-server}
      - PORT=${PORT:-5000}
      - HOST=${HOST:-0.0.0.0}
      # Default: config.yml bundled with the image (Hugging Face model, auto-downloaded).
      # Set to /models/config.yml to use your own config from a mounted models directory.
      - CONFIG_FILE=${CONFIG_FILE:-/llmaix/config.yml}
      - GPU=${GPU:-ALL}
      # Hugging Face token for gated/private repos used via hf_repo in config.yml (optional).
      - HF_TOKEN=${HF_TOKEN:-}
      - LLAMACPP_PORT=${LLAMACPP_PORT:-2929}
      - DEBUG=${DEBUG:-false}
      - MODE=${MODE:-choice}
      - VERBOSE_LLAMA=${VERBOSE_LLAMA:-false}
      # Set a password here if you want to protect the API (e.g. - PASSWORD:-llmaixpassword). If you set a password, the username is llmaix
      - PASSWORD=${PASSWORD:-}
      # Adjust the API URL and Key (e.g. - API_URL:-http://localhost:5000 - API_KEY:-someapikey)
      - API_URL=${API_URL:-}
      - API_KEY=${API_KEY:-}
      - ONLY_API=${ONLY_API:-false} # if true, a local model config will not be checked

volumes:
  llmaix_models: