interface/docker-compose.yml

services:
  # 1. LLM Generation Engine
  cachyllama:
    build:
      context: .
      dockerfile: Dockerfile
      args:
        CACHYLLAMA_COMMIT: ${CACHYLLAMA_COMMIT:-main}
    container_name: cachyllama-engine
    restart: unless-stopped
    devices:
      - /dev/dri:/dev/dri
    group_add:
      - video
      - render
    environment:
      - GGML_VK_FA_KV_CONTIG=1
      - GGML_VK_CONCAT_TRANSPOSE=1
    volumes:
      - ./data/models:/data/models:ro
      - ./data/ssd_cache:/data/ssd_cache:rw
    command:
      - --model
      - /data/models/${LLM_MODEL_FILE}
      - --host
      - 0.0.0.0
      - --port
      - "8080"
      - -c
      - ${LLM_CONTEXT_SIZE:-8192}
      - -ngl
      - ${LLM_GPU_LAYERS:-99}
      - --jinja
      - --temp
      - "${LLM_TEMP:-1.0}"
      - --top-p
      - "${LLM_TOP_P:-0.95}"
      - --top-k
      - "${LLM_TOP_K:-20}"
      - --cache-ssd
      - /data/ssd_cache
      - --cache-ssd-checkpoints
      - ${LLM_CACHE_SSD_CHECKPOINTS:-64}
      - --cache-ssd-hot-window
      - ${LLM_CACHE_SSD_HOT_WINDOW:-16384}
      - --cache-ssd-warm-window
      - ${LLM_CACHE_SSD_WARM_WINDOW:-32768}
      - --cache-ssd-system-prompts
      - ${LLM_CACHE_SYSTEM_PROMPTS:-8}
      - --cache-ssd-system-max-days
      - ${LLM_CACHE_SYSTEM_MAX_DAYS:-30}
      - --cache-ram
      - ${LLM_CACHE_RAM:-8192}
    ports:
      - "${LLM_HOST_PORT:-8080}:8080"
    healthcheck:
      test: ["CMD", "curl", "-fsS", "http://localhost:8080/health"]
      interval: 30s
      timeout: 5s
      retries: 3
      start_period: 30s
    networks:
      - internal

  # 2. Embedding Microservice
  harrier-embedder:
    build:
      context: .
      dockerfile: Dockerfile
      args:
        CACHYLLAMA_COMMIT: ${CACHYLLAMA_COMMIT:-main}
    container_name: embedder-engine
    restart: unless-stopped
    volumes:
      - ./data/models:/data/models:ro
    command:
      - --model
      - /data/models/${EMBED_MODEL_FILE}
      - --host
      - 0.0.0.0
      - --port
      - "8081"
      - --embedding
      - --pooling
      - ${EMBED_POOLING:-last}
      - -c
      - ${EMBED_CONTEXT_SIZE:-4096}
      - -ngl
      - "0"
      - -ub
      - "2048"
    ports:
      - "${EMBED_HOST_PORT:-8081}:8081"
    healthcheck:
      test: ["CMD", "curl", "-fsS", "http://localhost:8081/health"]
      interval: 30s
      timeout: 5s
      retries: 3
      start_period: 30s
    networks:
      - internal

  # 3. Redis Vector Store (RediSearch + RedisJSON for semantic caching)
  redis-stack:
    image: redis/redis-stack-server:7.4.0-v3
    container_name: redis-vector-store
    restart: unless-stopped
    environment:
      # requirepass gates access to the cache (semantic-cache prompts are
      # privacy-sensitive). REDIS_PASSWORD is sourced from .env.
      - REDIS_ARGS=--save 60 1 --appendonly yes --requirepass ${REDIS_PASSWORD}
    volumes:
      - ./data/redis:/data
    # NOTE: Redis is intentionally NOT published to the host. The LiteLLM
    # gateway reaches it over the `internal` docker network. To debug
    # locally, run `docker exec -it redis-vector-store redis-cli -a $REDIS_PASSWORD`.
    expose:
      - "6379"
    healthcheck:
      test: ["CMD", "sh", "-c", "redis-cli -a \"$$REDIS_PASSWORD\" ping | grep -q PONG"]
      interval: 30s
      timeout: 5s
      retries: 3
      start_period: 10s
    networks:
      - internal

  # 4. LiteLLM Proxy Gateway
  litellm:
    build:
      context: .
      dockerfile: Dockerfile.litellm
    container_name: litellm-proxy
    restart: unless-stopped
    depends_on:
      cachyllama:
        condition: service_healthy
      harrier-embedder:
        condition: service_healthy
      redis-stack:
        condition: service_healthy
    volumes:
      - ./litellm-config.yaml:/app/config.yaml:ro
    env_file:
      - .env
    environment:
      - REDIS_HOST=redis-stack
      - REDIS_PORT=6379
    command: ["--config", "/app/config.yaml", "--port", "4000"]
    ports:
      - "4000:4000"
    healthcheck:
      test: ["CMD", "curl", "-fsS", "http://localhost:4000/health/liveliness"]
      interval: 30s
      timeout: 5s
      retries: 3
      start_period: 20s
    networks:
      - internal
      - edge

networks:
  internal:
    driver: bridge
  edge:
    driver: bridge