kxkm_clown/docker-compose.yml

# ---------------------------------------------------------------------------
# KXKM_Clown — Docker Compose
# ---------------------------------------------------------------------------
# Usage:
#   docker compose up -d                          # postgres only
#   docker compose --profile v1 up -d             # V1 app + postgres
#   docker compose --profile v2 up -d             # V2 api + worker + postgres
#   docker compose --profile v1 --profile v2 up -d     # V1 + V2 + postgres
#   docker compose --profile ollama up -d          # include Ollama-compatible embeddings backend
#
# By default, the primary LLM runtime is expected on the host (LLM_URL, port 11434).
# OLLAMA_URL is retained for embeddings/RAG compatibility.
# ---------------------------------------------------------------------------

x-common-env: &common-env
  LLM_URL: "${LLM_URL:-http://host.docker.internal:11434}"
  LLM_MODEL: "${LLM_MODEL:-qwen-14b-awq}"
  OLLAMA_URL: "${OLLAMA_URL:-http://host.docker.internal:11434}"
  DATABASE_URL: postgres://kxkm:kxkm@postgres:5432/kxkm_clown
  NODE_ENV: production

x-extra-hosts: &extra-hosts
  - "host.docker.internal:host-gateway"

services:
  # -------------------------------------------------------------------------
  # V1 — main Express + WebSocket server
  # -------------------------------------------------------------------------
  app:
    build: .
    restart: unless-stopped
    profiles: [v1]
    extra_hosts: *extra-hosts
    ports:
      - "${APP_PORT:-3333}:3333"
    environment:
      <<: *common-env
      PORT: "3333"
      ADMIN_BOOTSTRAP_TOKEN: "${ADMIN_BOOTSTRAP_TOKEN:-}"
      ADMIN_TOKEN: "${ADMIN_TOKEN:-kxkm}"
      ADMIN_ALLOWED_SUBNETS: "${ADMIN_ALLOWED_SUBNETS:-}"
      WEB_SEARCH_API_BASE: "${WEB_SEARCH_API_BASE:-}"
      OWNER_NICK: "${OWNER_NICK:-}"
      MAX_GENERAL_RESPONDERS: "${MAX_GENERAL_RESPONDERS:-4}"
      CHAT_PAUSED: "${CHAT_PAUSED:-0}"
    volumes:
      - app-data:/app/data
    depends_on:
      postgres:
        condition: service_healthy
    healthcheck:
      test: ["CMD-SHELL", "node -e \"fetch('http://localhost:3333/api/ping').then(r=>{process.exit(r.ok?0:1)}).catch(()=>process.exit(1))\""]
      interval: 15s
      timeout: 5s
      retries: 3
      start_period: 10s

  # -------------------------------------------------------------------------
  # V2 — API server (Express, TypeScript) — main chat interface
  # Serves both the REST/WebSocket API and the React web build.
  # -------------------------------------------------------------------------
  api:
    build: .
    restart: unless-stopped
    profiles: [v2]
    network_mode: host
    command: ["node", "apps/api/dist/server.js"]
    volumes:
      - ./data:/app/data
      - ./scripts:/app/scripts:ro
      - /home/kxkm/openDAW/packages/app/studio/dist:/app/daw:ro
    environment:
      LLM_URL: "${LLM_URL:-http://localhost:11434}"
      LLM_MODEL: "${LLM_MODEL:-qwen-14b-awq}"
      OLLAMA_URL: "http://localhost:11434"
      DATABASE_URL: "postgres://kxkm:kxkm@localhost:5432/kxkm_clown"
      NODE_ENV: production
      V2_API_PORT: "${API_PORT:-3333}"
      WEB_DIST_PATH: "/app/apps/web/dist"
      DAW_DIST_PATH: "/app/daw"
      ADMIN_BOOTSTRAP_TOKEN: "${ADMIN_BOOTSTRAP_TOKEN:-}"
      ADMIN_TOKEN: "${ADMIN_TOKEN:-kxkm}"
      ADMIN_ALLOWED_SUBNETS: "${ADMIN_ALLOWED_SUBNETS:-}"
      TTS_ENABLED: "1"
      KOKORO_URL: "http://127.0.0.1:9201"
      PYTHON_BIN: "python3"
      SCRIPTS_DIR: "/app/scripts"
      PIPER_VOICE_DIR: "/app/data/piper-voices"
      COQUI_TOS_AGREED: "1"
      VISION_MODEL: "qwen3-vl:8b"
      LIGHTRAG_URL: "http://localhost:9621"
      TTS_URL: "http://localhost:9100"
      DOCLING_URL: "http://localhost:9400"
      RERANKER_URL: "http://localhost:9500"
      RAG_CHUNK_SIZE: "${RAG_CHUNK_SIZE:-500}"
      RAG_MIN_SIMILARITY: "${RAG_MIN_SIMILARITY:-0.3}"
      RAG_MAX_RESULTS: "${RAG_MAX_RESULTS:-3}"
      RAG_EMBEDDING_MODEL: "${RAG_EMBEDDING_MODEL:-nomic-embed-text}"
    depends_on:
      postgres:
        condition: service_healthy
    healthcheck:
      test: ["CMD-SHELL", "node -e \"fetch('http://localhost:3333/api/v2/health').then(r=>{process.exit(r.ok?0:1)}).catch(()=>process.exit(1))\""]
      interval: 15s
      timeout: 5s
      retries: 3
      start_period: 10s

  # -------------------------------------------------------------------------
  # V2 — Worker (poll loop, GPU access for training)
  # -------------------------------------------------------------------------
  worker:
    build: .
    restart: unless-stopped
    profiles: [v2]
    network_mode: host
    command: ["node", "apps/worker/dist/index.js"]
    environment:
      LLM_URL: "${LLM_URL:-http://localhost:11434}"
      LLM_MODEL: "${LLM_MODEL:-qwen-14b-awq}"
      OLLAMA_URL: "http://localhost:11434"
      DATABASE_URL: "postgres://kxkm:kxkm@localhost:5432/kxkm_clown"
      NODE_ENV: production
      PYTHON_BIN: "${PYTHON_BIN:-/home/kxkm/venv/bin/python3}"
      SCRIPTS_DIR: "${SCRIPTS_DIR:-/app/scripts}"
      TRAINING_TIMEOUT_MS: "${TRAINING_TIMEOUT_MS:-3600000}"
    volumes:
      - /home/kxkm/venv:/home/kxkm/venv:ro
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: all
              capabilities: [gpu]
    depends_on:
      postgres:
        condition: service_healthy

  # -------------------------------------------------------------------------
  # PostgreSQL 16
  # -------------------------------------------------------------------------
  postgres:
    image: postgres:16-alpine
    restart: unless-stopped
    environment:
      POSTGRES_USER: kxkm
      POSTGRES_PASSWORD: kxkm
      POSTGRES_DB: kxkm_clown
    volumes:
      - pg-data:/var/lib/postgresql/data
    ports:
      - "${PG_PORT:-5432}:5432"
    healthcheck:
      test: ["CMD-SHELL", "pg_isready -U kxkm -d kxkm_clown"]
      interval: 5s
      timeout: 3s
      retries: 5
      start_period: 5s

  # -------------------------------------------------------------------------
  # Ollama-compatible embeddings backend (optional, use --profile ollama)
  # -------------------------------------------------------------------------
  ollama:
    image: ollama/ollama:latest
    restart: unless-stopped
    profiles: [ollama]
    ports:
      - "${OLLAMA_PORT:-11434}:11434"
    volumes:
      - ollama-data:/root/.ollama
    # GPU passthrough — uncomment if NVIDIA GPU available:
    # deploy:
    #   resources:
    #     reservations:
    #       devices:
    #         - driver: nvidia
    #           count: all
    #           capabilities: [gpu]

  # -------------------------------------------------------------------------
  # SearXNG — self-hosted privacy search engine
  # Replaces DuckDuckGo Lite scraping with structured JSON API.
  # Usage: set WEB_SEARCH_API_BASE=http://localhost:8080/search in .env
  # -------------------------------------------------------------------------
  searxng:
    image: searxng/searxng:latest
    restart: unless-stopped
    profiles: [v2]
    ports:
      - "${SEARXNG_PORT:-8080}:8080"
    volumes:
      - ./ops/v2/searxng:/etc/searxng:ro
    environment:
      SEARXNG_BASE_URL: "http://localhost:8080/"
    healthcheck:
      test: ["CMD", "wget", "-q", "--spider", "http://localhost:8080/"]
      interval: 30s
      timeout: 5s
      retries: 3
      start_period: 10s

  # -------------------------------------------------------------------------
  # Discord Pharmacius Bot — bridges KXKM chat ↔ Discord
  # Usage: DISCORD_BOT_TOKEN=xxx DISCORD_CHANNEL_ID=yyy docker compose --profile discord up -d
  # -------------------------------------------------------------------------
  discord-bot:
    build: .
    restart: unless-stopped
    profiles: [discord]
    network_mode: host
    command: ["node", "scripts/discord-pharmacius.js"]
    environment:
      DISCORD_BOT_TOKEN: "${DISCORD_BOT_TOKEN:-}"
      DISCORD_CHANNEL_ID: "${DISCORD_CHANNEL_ID:-}"
      KXKM_WS_URL: "ws://localhost:3333/ws"
    depends_on:
      api:
        condition: service_healthy

  # -------------------------------------------------------------------------
  # Discord Voice Bot — STT → Personas → TTS in voice channels
  # Usage: DISCORD_BOT_TOKEN=xxx DISCORD_VOICE_CHANNEL=yyy docker compose --profile discord-voice up -d
  # -------------------------------------------------------------------------
  discord-voice:
    build: .
    restart: unless-stopped
    profiles: [discord-voice]
    network_mode: host
    command: ["node", "scripts/discord-voice.js"]
    environment:
      DISCORD_BOT_TOKEN: "${DISCORD_BOT_TOKEN:-}"
      DISCORD_VOICE_CHANNEL: "${DISCORD_VOICE_CHANNEL_ID_2:-}"
      KXKM_WS_URL: "ws://localhost:3333/ws"
      PYTHON_BIN: "${PYTHON_BIN:-/home/kxkm/venv/bin/python3}"
      SCRIPTS_DIR: "/app/scripts"
    volumes:
      - /home/kxkm/venv:/home/kxkm/venv:ro
    depends_on:
      api:
        condition: service_healthy

  # -------------------------------------------------------------------------
  # Chatterbox TTS — GPU-accelerated voice synthesis (Docker)
  # Usage: docker compose --profile v2 up -d
  # API: POST /tts, POST /v1/audio/speech
  # -------------------------------------------------------------------------
  chatterbox:
    image: ghcr.io/devnen/chatterbox-tts-server:latest
    restart: unless-stopped
    profiles: [v2]
    ports:
      - "${CHATTERBOX_PORT:-9200}:8004"
    volumes:
      - ./data/voice-samples:/app/voices:ro
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: 1
              capabilities: [gpu]
    healthcheck:
      test: ["CMD-SHELL", "python3 -c \"import urllib.request; urllib.request.urlopen('http://localhost:8004/get_predefined_voices')\""]
      interval: 30s
      timeout: 10s
      retries: 3
      start_period: 60s

  # -------------------------------------------------------------------------
  # TTS Sidecar — Proxy to Chatterbox + Piper fallback
  # Runs on host network, proxies /synthesize to Chatterbox :9200
  # -------------------------------------------------------------------------
  tts-sidecar:
    build: .
    restart: unless-stopped
    profiles: [v2]
    network_mode: host
    command: ["python3", "scripts/tts-server.py", "--port", "9100", "--backend", "chatterbox-remote"]
    environment:
      CHATTERBOX_URL: "http://127.0.0.1:9200"
      PIPER_VOICE_DIR: "/app/data/piper-voices"
      KXKM_VOICE_SAMPLES_DIR: "/app/data/voice-samples"
    volumes:
      - ./data:/app/data
      - ./scripts:/app/scripts:ro
      - /home/kxkm/openDAW/packages/app/studio/dist:/app/daw:ro
    depends_on:
      chatterbox:
        condition: service_healthy

  # -------------------------------------------------------------------------
  # Docling Serve — Document parsing (PDF, DOCX, PPTX → structured text)
  # API: POST /v1/convert/source
  # UI:  http://localhost:9400/ui (if DOCLING_SERVE_ENABLE_UI=1)
  # -------------------------------------------------------------------------
  docling:
    image: ghcr.io/docling-project/docling-serve:latest
    restart: unless-stopped
    profiles: [v2]
    ports:
      - "9400:5001"
    environment:
      DOCLING_SERVE_ENABLE_UI: "1"
    healthcheck:
      test: ["CMD-SHELL", "python3 -c \"import urllib.request; urllib.request.urlopen('http://localhost:5001/health')\""]
      interval: 30s
      timeout: 10s
      retries: 3
      start_period: 120s

  # -------------------------------------------------------------------------
  # LightRAG — Graph RAG server
  # API: POST /query, POST /documents/text, GET /health
  # Web UI: http://localhost:9621
  # -------------------------------------------------------------------------
  lightrag:
    build:
      context: .
      dockerfile_inline: |
        FROM python:3.12-slim
        RUN pip install --no-cache-dir 'lightrag-hku[api]'
        EXPOSE 9621
        CMD ["lightrag-server", "--host", "0.0.0.0", "--port", "9621"]
    restart: unless-stopped
    profiles: [v2]
    network_mode: host
    environment:
      LLM_MODEL: "qwen3:8b"
      EMBEDDING_MODEL: "nomic-embed-text"
      EMBEDDING_DIM: "768"
      OLLAMA_HOST: "${OLLAMA_URL:-http://localhost:11434}"
      LLM_BINDING: ollama
      EMBEDDING_BINDING: ollama
      RAG_DIR: "/data/lightrag"
    command: >
      lightrag-server
        --host 0.0.0.0
        --port 9621
        --working-dir /data/lightrag
        --llm-binding ollama
        --embedding-binding ollama
    volumes:
      - ./data/lightrag:/data/lightrag
    healthcheck:
      test: ["CMD-SHELL", "python3 -c \"import urllib.request; urllib.request.urlopen('http://localhost:9621/health')\""]
      interval: 30s
      timeout: 10s
      retries: 3
      start_period: 30s

  # -------------------------------------------------------------------------
  # Prometheus — metrics collection
  # Usage: docker compose --profile monitoring up -d
  # -------------------------------------------------------------------------
  prometheus:
    image: prom/prometheus:latest
    restart: unless-stopped
    profiles: [monitoring]
    extra_hosts: *extra-hosts
    ports:
      - "9090:9090"
    volumes:
      - ./ops/monitoring/prometheus.yml:/etc/prometheus/prometheus.yml:ro
      - prometheus-data:/prometheus
    command:
      - "--config.file=/etc/prometheus/prometheus.yml"
      - "--storage.tsdb.retention.time=30d"
      - "--web.enable-lifecycle"

  # -------------------------------------------------------------------------
  # Grafana — dashboards & visualization
  # Usage: docker compose --profile monitoring up -d
  # Default login: admin / admin
  # -------------------------------------------------------------------------
  grafana:
    image: grafana/grafana:latest
    restart: unless-stopped
    profiles: [monitoring]
    ports:
      - "3000:3000"
    environment:
      GF_SECURITY_ADMIN_PASSWORD: "${GRAFANA_PASSWORD:-kxkm}"
      GF_USERS_ALLOW_SIGN_UP: "false"
    volumes:
      - grafana-data:/var/lib/grafana
      - ./ops/monitoring/grafana/provisioning:/etc/grafana/provisioning:ro
      - ./ops/monitoring/grafana/dashboards:/var/lib/grafana/dashboards:ro
    depends_on:
      - prometheus

  # --- Text Embeddings Inference (dedicated embedding server) -----------------
  # Usage: docker compose --profile embeddings up -d
  # Exposes OpenAI-compatible /v1/embeddings on port 9500
  # -------------------------------------------------------------------------
  tei:
    image: ghcr.io/huggingface/text-embeddings-inference:cpu-1.6
    container_name: kxkm-tei
    restart: unless-stopped
    profiles: [v2, embeddings]
    ports:
      - "9500:80"
    volumes:
      - tei-models:/data
    command: --model-id BAAI/bge-m3 --port 80

volumes:
  app-data:
  pg-data:
  ollama-data:
  tei-models:
  prometheus-data:
  grafana-data: