services:
  vllm:
    image: vllm/vllm-openai:latest
    container_name: cortex-vllm
    restart: unless-stopped
    runtime: nvidia
    ipc: host
    ports:
      - "8005:8005"
    environment:
      - NVIDIA_VISIBLE_DEVICES=0,1
      - HUGGING_FACE_HUB_TOKEN=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
      - PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True
    volumes:
      - /opt/vllm/huggingface:/root/.cache/huggingface
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: 2
              capabilities: [gpu]
    command:
       - --model=Qwen/Qwen3-14B-AWQ
       - --port=8005
       - --quantization=awq
       - --tensor-parallel-size=2
       - --gpu-memory-utilization=0.86
       - --max-model-len=60000
       - --hf-overrides={"rope_scaling":{"rope_type":"yarn","factor":2.0,"original_max_position_embeddings":32768}}
       - --max-num-seqs=4
       - --served-model-name=qwen3-14b-awq
       - --enable-auto-tool-choice
       - --tool-call-parser=hermes
    healthcheck:
      test: ["CMD", "curl", "-f", "http://localhost:8005/health"]
      interval: 30s
      timeout: 10s
      retries: 5
      start_period: 180s
    networks:
      - cortex-net

  hermes-gateway:
    extra_hosts:
      - "host.docker.internal:host-gateway"
    image: nousresearch/hermes-agent:latest
    container_name: cortex-hermes-gateway
    restart: unless-stopped
    user: "0:0"
    privileged: true
    depends_on:
      vllm:
        condition: service_healthy
    command: ["gateway", "run"]
    volumes:
      - /opt/hermes:/opt/data
      - /opt/hermes/.ssh:/root/.ssh
      - /var/run/docker.sock:/var/run/docker.sock
    environment:
      - OPENAI_BASE_URL=http://vllm:8005/v1
      - OPENAI_API_KEY=sk-dummy-key-1234567890
      - HERMES_WRITE_SAFE_ROOT=/
    networks:
      - cortex-net

  hermes-dashboard:
    user: "0:0"
    image: nousresearch/hermes-agent:latest
    container_name: cortex-hermes-dashboard
    restart: unless-stopped
    depends_on:
      - hermes-gateway
    command: ["dashboard", "--host", "0.0.0.0", "--no-open"]
    ports:
      - "9119:9119"
    volumes:
      - /opt/hermes:/opt/data
      - /opt/hermes/.ssh:/root/.ssh   
    environment:
      - GATEWAY_HEALTH_URL=http://cortex-hermes-gateway:8642
      - HERMES_WRITE_SAFE_ROOT=/
    networks:
      - cortex-net

networks:
  cortex-net:
    driver: bridge