docker-compose.gpu.yml

# docker-compose.gpu.yml
# GPU-enabled version (requires NVIDIA GPU + nvidia-docker)
services:
  # ============================================================
  # WEB APPLICATION (Required)
  # ============================================================
  web:
    build: .
    container_name: flai-web
    ports:
      - "5000:5000"
    depends_on:
      - redis
      - postgres
      - llama-swap
    env_file:
      - .env
    volumes:
      # Mount application data directory
      - ./data:/app/data
      - ./.env:/app/.env:ro
      # Mount models directory for GGUF metadata reading
      - ./services/llamacpp/models:/models:ro
      # Mount llama-swap config
      - ./llama-swap-config:/config
      # Mount docker socket for GPU detection
      - /var/run/docker.sock:/var/run/docker.sock
      # Persistent code mounts (survive restart)
      - ./app/llamacpp_client.py:/app/app/llamacpp_client.py
      - ./app/llama_swap_config.py:/app/app/llama_swap_config.py
      - ./app/queue.py:/app/app/queue.py
      - ./app/resource_manager.py:/app/app/resource_manager.py
      - ./app/utils.py:/app/app/utils.py
      - ./app/database.py:/app/app/database.py
      - ./app/db.py:/app/app/db.py
      - ./app/routes/admin.py:/app/app/routes/admin.py
      - ./app/routes/queue.py:/app/app/routes/queue.py
      - ./app/routes/messages.py:/app/app/routes/messages.py
      - ./app/static/js/chat-init.js:/app/app/static/js/chat-init.js
      - ./app/static/js/chat-recording.js:/app/app/static/js/chat-recording.js
      - ./app/static/js/chat-constants.js:/app/app/static/js/chat-constants.js
      - ./app/__init__.py:/app/app/__init__.py
      - ./modules/video.py:/app/modules/video.py
      - ./modules/sd_cpp.py:/app/modules/sd_cpp.py
      - ./app/static/js/events.js:/app/app/static/js/events.js
      - ./app/static/js/admin-models.js:/app/app/static/js/admin-models.js
      - ./app/static/js/chat-queue.js:/app/app/static/js/chat-queue.js
      - ./app/static/js/chat-messages.js:/app/app/static/js/chat-messages.js
      - ./prompts/en/rag.template:/app/prompts/en/rag.template
      - ./prompts/ru/rag.template:/app/prompts/ru/rag.template
      - ./prompts/en/reasoning.template:/app/prompts/en/reasoning.template
      - ./prompts/ru/reasoning.template:/app/prompts/ru/reasoning.template
      - ./prompts/en/base_text.template:/app/prompts/en/base_text.template
      - ./prompts/ru/base_text.template:/app/prompts/ru/base_text.template
    runtime: nvidia
    group_add:
      - "988"  # Docker socket group — needed for `docker restart` from inside container
    environment:
      - REDIS_URL=redis://redis:6379/0
      - NVIDIA_VISIBLE_DEVICES=all
      - NVIDIA_DRIVER_CAPABILITIES=compute,utility
    networks:
      - flai_network
    restart: unless-stopped
    healthcheck:
      test: ["CMD", "python3", "-c", "import urllib.request; urllib.request.urlopen('http://localhost:5000/health')"]
      interval: 30s
      timeout: 10s
      retries: 3
      start_period: 10s
    # Resource limits for production
    deploy:
      resources:
        limits:
          cpus: '4.0'
          memory: 4G
        reservations:
          cpus: '0.5'
          memory: 512M
    # Logging configuration
    logging:
      driver: "json-file"
      options:
        max-size: "10m"
        max-file: "3"

  # ============================================================
  # REDIS (Required - Request Queue)
  # ============================================================
  redis:
    image: redis:8.4.2-alpine
    container_name: flai-redis
    volumes:
      - redis_data:/data
    command: redis-server --appendonly yes
    networks:
      - flai_network
    restart: unless-stopped
    healthcheck:
      test: ["CMD", "redis-cli", "ping"]
      interval: 30s
      timeout: 10s
      retries: 3
      start_period: 10s
    deploy:
      resources:
        limits:
          cpus: '1.0'
          memory: 256M
        reservations:
          cpus: '0.25'
          memory: 64M
    logging:
      driver: "json-file"
      options:
        max-size: "10m"
        max-file: "3"

  # ============================================================
  # LLAMA-SWAP (Required - LLM Proxy, router mode)
  # ============================================================
  llama-swap:
    image: ghcr.io/mostlygeek/llama-swap:cuda
    container_name: flai-llamaswap
    volumes:
      - ./services/llamacpp/models:/models:ro
      - ./llama-swap-config:/config
    environment:
      - NVIDIA_VISIBLE_DEVICES=all
      - NVIDIA_DRIVER_CAPABILITIES=compute,utility
    runtime: nvidia
    networks:
      - flai_network
    restart: unless-stopped
    command: -config /config/llama-swap.yaml -watch-config
    healthcheck:
      test: ["CMD-SHELL", "curl -f http://localhost:8080/health || exit 1"]
      interval: 30s
      timeout: 10s
      retries: 3
      start_period: 60s
    logging:
      driver: "json-file"
      options:
        max-size: "10m"
        max-file: "3"

  # ============================================================
  # STABLE-DIFFUSION.CPP (Optional - Image Generation)
  # ============================================================
  sd_cpp:
    build:
      context: ./services/sd_cpp
      dockerfile: Dockerfile.sd_cpp
    container_name: flai-sd
    volumes:
      - ./services/sd_cpp/models/diffusion_models:/app/models/diffusion_models
      - ./services/sd_cpp/models/vae:/app/models/vae
      - ./services/sd_cpp/models/text_encoders:/app/models/text_encoders
      - ./services/sd_cpp/outputs:/app/outputs
      - ./services/sd_cpp/sd_wrapper.py:/app/sd_wrapper.py:ro
    environment:
      - NVIDIA_VISIBLE_DEVICES=all
      - NVIDIA_DRIVER_CAPABILITIES=compute,utility
    runtime: nvidia
    deploy:
      resources:
        limits:
          cpus: '8.0'
          memory: 10G
        reservations:
          cpus: '2.0'
          memory: 4G
    networks:
      - flai_network
    restart: unless-stopped
    # Run sd-wrapper HTTP API on port 7861 (calls sd-cli internally)
    entrypoint: ["python3", "/app/sd_wrapper.py"]
    healthcheck:
      test: ["CMD", "curl", "-f", "http://localhost:7861/health"]
      interval: 30s
      timeout: 10s
      retries: 3
      start_period: 10s
    profiles:
      - with-image-gen
    logging:
      driver: "json-file"
      options:
        max-size: "10m"
        max-file: "3"

  # ============================================================
  # LTX-VIDEO (Optional - Video Generation)
  # ============================================================
  ltxvideo:
    build:
      context: ./services/ltx_video
      dockerfile: Dockerfile.ltx_video
    container_name: flai-ltxvideo
    volumes:
      - ./services/ltx_video/models:/app/models:rw
      - ./services/llamacpp/models:/models:ro
    environment:
      - NVIDIA_VISIBLE_DEVICES=all
      - NVIDIA_DRIVER_CAPABILITIES=compute,utility
      - LTX_MODELS_DIR=/app/models
      - LTX_DEFAULT_MODEL=ltxv-2b-0.9.8-distilled.safetensors
      - LTX_DEFAULT_UPSCALER=ltxv-spatial-upscaler-0.9.8.safetensors
      - LTX_PIPELINE_CONFIG=ltxv-2b-0.9.8-distilled.yaml
      - LTX_TEXT_ENCODER_PATH=/app/models/t5_encoder
    runtime: nvidia
    networks:
      - flai_network
    restart: unless-stopped
    healthcheck:
      test: ["CMD", "curl", "-f", "http://localhost:7872/health"]
      interval: 30s
      timeout: 10s
      retries: 3
      start_period: 180s
    deploy:
      resources:
        limits:
          cpus: '8.0'
          memory: 16G
        reservations:
          cpus: '2.0'
          memory: 8G
    profiles:
      - with-video
    logging:
      driver: "json-file"
      options:
        max-size: "10m"
        max-file: "3"

  # ============================================================
  # WHISPER (Optional - Speech Recognition)
  # ============================================================
  whisper:
    image: onerahmet/openai-whisper-asr-webservice:v1.9.1
    container_name: flai-whisper
    environment:
      - ASR_MODEL=medium
      - ASR_ENGINE=faster_whisper
      - ASR_DEVICE=cpu
    volumes:
      - ./data/hf-cache/hub:/root/.cache/whisper
    networks:
      - flai_network
    restart: unless-stopped
    healthcheck:
      test: ["CMD", "curl", "-f", "http://localhost:9000/docs"]
      interval: 15s
      timeout: 5s
      retries: 3
      start_period: 30s
    deploy:
      resources:
        limits:
          cpus: '2.0'
          memory: 2G
        reservations:
          cpus: '0.5'
          memory: 1G
    logging:
      driver: "json-file"
      options:
        max-size: "10m"
        max-file: "3"
    profiles:
      - with-voice

  # ============================================================
  # PIPER TTS (Optional - Speech Synthesis)
  # ============================================================
  piper:
    build:
      context: ./services/piper
      dockerfile: Dockerfile.piper
    container_name: flai-piper
    volumes:
      - ./services/piper/piper_models:/app/models
    environment:
      - PIPER_MODEL_DIR=/app/models
    networks:
      - flai_network
    restart: unless-stopped
    healthcheck:
      test: ["CMD-SHELL", "python3 -c \"import urllib.request; urllib.request.urlopen('http://localhost:8888/health')\" || exit 1"]
      interval: 30s
      timeout: 10s
      retries: 3
      start_period: 30s
    deploy:
      resources:
        limits:
          cpus: '2.0'
          memory: 512M
        reservations:
          cpus: '0.5'
          memory: 256M
    logging:
      driver: "json-file"
      options:
        max-size: "10m"
        max-file: "3"
    profiles:
      - with-voice

  # ============================================================
  # SUPERLOCALMEMORY (Optional - Long-term memory)
  # ============================================================
  superlocalmemory:
    build:
      context: ./services/superlocalmemory
      dockerfile: Dockerfile.slm
    container_name: flai-slm
    environment:
      - SLM_MODE=a
      - SLM_PORT=8765
      - SLM_HOST=0.0.0.0
    volumes:
      - slm_data:/root/.superlocalmemory
      - ./data/slm/:/app/data/slm/
    networks:
      - flai_network
    restart: unless-stopped
    healthcheck:
      test: ["CMD", "python3", "-c", "import urllib.request; urllib.request.urlopen('http://localhost:8766/health', timeout=5)"]
      interval: 30s
      timeout: 10s
      retries: 3
      start_period: 45s
    deploy:
      resources:
        limits:
          cpus: '2.0'
          memory: 1G
        reservations:
          cpus: '0.5'
          memory: 512M
    profiles:
      - with-slm

  # ============================================================
  # QDRANT (Optional - Vector DB for RAG)
  # ============================================================
  qdrant:
    image: qdrant/qdrant:v1.12.1
    container_name: flai-qdrant
    volumes:
      - qdrant_data:/qdrant/storage
    environment:
      - QDRANT__SERVICE__API_KEY=${QDRANT_API_KEY}
      - QDRANT__SERVICE__ENABLE_TLS=0
    networks:
      - flai_network
    restart: unless-stopped
    healthcheck:
      test: ["CMD-SHELL", "bash -c 'echo > /dev/tcp/localhost/6333' || exit 1"]
      interval: 30s
      timeout: 10s
      retries: 3
      start_period: 30s
    deploy:
      resources:
        limits:
          cpus: '2.0'
          memory: 1G
        reservations:
          cpus: '0.5'
          memory: 512M
    logging:
      driver: "json-file"
      options:
        max-size: "10m"
        max-file: "3"
    profiles:
      - with-rag

  # ============================================================
  # POSTGRESQL (Required - Application DB)
  # ============================================================
  postgres:
    image: postgres:18-alpine
    container_name: flai-postgres
    volumes:
      - postgres_data:/var/lib/postgresql
    environment:
      - POSTGRES_USER=${POSTGRES_USER:-flai}
      - POSTGRES_PASSWORD=${POSTGRES_PASSWORD:-flai_password}
      - POSTGRES_DB=${POSTGRES_DB:-flai}
    networks:
      - flai_network
    restart: unless-stopped
    healthcheck:
      test: ["CMD-SHELL", "pg_isready -U flai"]
      interval: 30s
      timeout: 10s
      retries: 3
      start_period: 30s
    deploy:
      resources:
        limits:
          cpus: '2.0'
          memory: 1G
        reservations:
          cpus: '0.5'
          memory: 256M

networks:
  flai_network:
    name: flai_network
    driver: bridge

volumes:
  redis_data:
  postgres_data:
  qdrant_data:
  slm_data: