# ODS - external Lemonade SDK runtime overlay
#
# Use when Lemonade is already installed and running on the host. This overlay
# keeps ODS's managed llama-server disabled and routes ODS services
# through LiteLLM, which calls the existing Lemonade service.
#
# This file is intentionally not standalone. It is selected by
# scripts/resolve-compose-stack.sh together with:
#
#   docker-compose.base.yml
#   docker-compose.cloud.yml
#   extensions/services/litellm/compose.yaml
#   docker-compose.lemonade-external.yml
#
# Validate the resolved stack with:
#
#   LEMONADE_EXTERNAL=true ODS_MODE=lemonade \
#     ./scripts/resolve-compose-stack.sh --ods-mode lemonade --gpu-backend amd --tier SH_LARGE

services:
  litellm:
    extra_hosts:
      - "host.docker.internal:host-gateway"

  open-webui:
    environment:
      OPENAI_API_BASE_URL: "http://litellm:4000/v1"
      OPENAI_API_KEY: "${LITELLM_KEY}"

  dashboard-api:
    environment:
      - GPU_BACKEND=${GPU_BACKEND:-amd}
      - LLM_BACKEND=lemonade
      - LLM_API_BASE_PATH=${LEMONADE_API_BASE_PATH:-/api/v1}
      - OLLAMA_URL=${LEMONADE_CONTAINER_BASE_URL:-http://host.docker.internal:13305}
      - LLM_URL=${LLM_API_URL:-http://litellm:4000}
      - ODS_TALK_HERMES_TIMEOUT=${ODS_TALK_HERMES_TIMEOUT:-900}
      - ODS_TALK_VISION_URL=${ODS_TALK_VISION_URL:-http://host.docker.internal:${AMD_INFERENCE_PORT:-13305}${LEMONADE_API_BASE_PATH:-/api/v1}}
      - AMD_INFERENCE_RUNTIME=${AMD_INFERENCE_RUNTIME:-lemonade}
      - AMD_INFERENCE_BACKEND=${AMD_INFERENCE_BACKEND:-auto}
      - AMD_INFERENCE_LOCATION=${AMD_INFERENCE_LOCATION:-host}
      - AMD_INFERENCE_PORT=${AMD_INFERENCE_PORT:-13305}
      - AMD_INFERENCE_SUPPORTED_BACKENDS=${AMD_INFERENCE_SUPPORTED_BACKENDS:-auto}
      - AMD_INFERENCE_RUNTIME_MODE=${AMD_INFERENCE_RUNTIME_MODE:-external-lemonade}
      - AMD_INFERENCE_MANAGED=${AMD_INFERENCE_MANAGED:-false}