docker-compose.yml

services:
  fleet-controller:
    build:
      context: ./controller
      args:
        CODEX_NPM_REFRESH_TOKEN: ${CODEX_NPM_REFRESH_TOKEN:-static}
    env_file:
      - ./runtime.ea.env
      - ./runtime.env
    container_name: fleet-controller
    restart: unless-stopped
    init: true
    user: "${LOCAL_UID:-1000}:${LOCAL_GID:-1000}"
    environment:
      APP_PORT: 8090
      FLEET_CONFIG_PATH: /app/config/fleet.yaml
      FLEET_ACCOUNTS_PATH: /app/config/accounts.yaml
      FLEET_DB_PATH: /var/lib/codex-fleet/fleet.db
      FLEET_CONTROLLER_HEARTBEAT_PATH: /var/lib/codex-fleet/controller-heartbeat.json
      FLEET_CONTROLLER_HEARTBEAT_MAX_AGE_SECONDS: "45"
      FLEET_CONTROLLER_HEALTH_TIMEOUT_SECONDS: "10"
      FLEET_CONTROLLER_HEALTH_ALLOW_HEARTBEAT_ONLY: "1"
      FLEET_LOG_DIR: /var/lib/codex-fleet/logs
      FLEET_CODEX_HOME_ROOT: /var/lib/codex-fleet/codex-homes
      FLEET_GH_HOSTS_PATH: /run/gh/hosts.yml
      FLEET_QUARTERMASTER_BASE_URL: http://fleet-quartermaster:8094
      TZ: UTC
    volumes:
      - ./state:/var/lib/codex-fleet
      - ./config:/app/config:ro
      - ./secrets:/run/secrets:ro
      - ${HOME:-/home/tibor}/.codex/auth.json:/run/secrets/chatgpt.auth.json:ro
      - ${HOME:-/home/tibor}/.config/gh:/run/gh:ro
      - ./:${FLEET_SELF_MOUNT_PATH:-/opt/codex-fleet}
      - /docker:/docker
    extra_hosts:
      - "host.docker.internal:host-gateway"
    networks:
      - fleet_net
    healthcheck:
      test: ["CMD-SHELL", "python /opt/codex-fleet/scripts/healthcheck_controller.py"]
      interval: 30s
      timeout: 15s
      retries: 3
      start_period: 20s

  fleet-studio:
    build:
      context: ./studio
      args:
        CODEX_NPM_REFRESH_TOKEN: ${CODEX_NPM_REFRESH_TOKEN:-static}
    env_file:
      - ./runtime.ea.env
      - ./runtime.env
    container_name: fleet-studio
    restart: unless-stopped
    init: true
    user: "${LOCAL_UID:-1000}:${LOCAL_GID:-1000}"
    environment:
      APP_PORT: 8091
      FLEET_CONFIG_PATH: /app/config/fleet.yaml
      FLEET_ACCOUNTS_PATH: /app/config/accounts.yaml
      FLEET_DB_PATH: /var/lib/codex-fleet/fleet.db
      FLEET_LOG_DIR: /var/lib/codex-fleet/studio-logs
      FLEET_CODEX_HOME_ROOT: /var/lib/codex-fleet/codex-homes
      FLEET_GH_HOSTS_PATH: /run/gh/hosts.yml
      TZ: UTC
    volumes:
      - ./state:/var/lib/codex-fleet
      - ./config:/app/config:ro
      - ./secrets:/run/secrets:ro
      - ${HOME:-/home/tibor}/.codex/auth.json:/run/secrets/chatgpt.auth.json:ro
      - ${HOME:-/home/tibor}/.config/gh:/run/gh:ro
      - ./:${FLEET_SELF_MOUNT_PATH:-/opt/codex-fleet}
      - /docker:/docker
    extra_hosts:
      - "host.docker.internal:host-gateway"
    networks:
      - fleet_net
    healthcheck:
      test: ["CMD-SHELL", "python - <<'PY'\nimport urllib.request\nurllib.request.urlopen('http://127.0.0.1:8091/health', timeout=5)\nPY"]
      interval: 30s
      timeout: 5s
      retries: 3
      start_period: 20s

  fleet-design-supervisor:
    build:
      context: ./controller
      args:
        CODEX_NPM_REFRESH_TOKEN: ${CODEX_NPM_REFRESH_TOKEN:-static}
    env_file:
      - ./runtime.ea.env
      - ./runtime.env
    container_name: fleet-design-supervisor
    restart: unless-stopped
    init: true
    user: "${LOCAL_UID:-1000}:${LOCAL_GID:-1000}"
    working_dir: ${FLEET_SELF_MOUNT_PATH:-/opt/codex-fleet}
    environment:
      CHUMMER_DESIGN_SUPERVISOR_STATE_ROOT: /var/lib/codex-fleet/chummer_design_supervisor
      CHUMMER_DESIGN_SUPERVISOR_CLEAR_LOCK_ON_BOOT: "1"
      CHUMMER_DESIGN_SUPERVISOR_HEALTH_MAX_AGE_SECONDS: "900"
      CHUMMER_DESIGN_SUPERVISOR_PREFER_CORE_BATCH_MODEL_ALIASES: "0"
      CHUMMER_DESIGN_SUPERVISOR_WATCHDOG_STARTUP_GRACE_SECONDS: "1800"
      CHUMMER_DESIGN_SUPERVISOR_LIZ_TRANSPORT_RECONNECT_STALL_SECONDS: "1800"
      CHUMMER_DESIGN_SUPERVISOR_STATUS_HELPER_LOOP_TRIP_THRESHOLD: "4"
      CHUMMER_DESIGN_SUPERVISOR_TASK_LOCAL_STATUS_LOOP_GRACE_SECONDS: "45"
      CHUMMER_DESIGN_SUPERVISOR_SELF_POLL_BLOCKED_TRIP_THRESHOLD: "0"
      CHUMMER_DESIGN_SUPERVISOR_ACCOUNT_ALIASES: ""
      CHUMMER_DESIGN_SUPERVISOR_DISABLE_OPENAI_ESCAPE: "1"
      CHUMMER_DESIGN_SUPERVISOR_OPENAI_ESCAPE_ACCOUNT_ALIASES: ""
      CHUMMER_DESIGN_SUPERVISOR_OPENAI_ESCAPE_ACCOUNT_OWNER_IDS: ""
      CHUMMER_DESIGN_SUPERVISOR_OPENAI_ESCAPE_MODELS: ""
      TZ: UTC
    volumes:
      - ./state:/var/lib/codex-fleet
      - ./config:/app/config:ro
      - ./secrets:/run/secrets:ro
      - ${HOME:-/home/tibor}/.codex/auth.json:/run/secrets/chatgpt.auth.json:ro
      - ${HOME:-/home/tibor}/.config/gh:/run/gh:ro
      - ./:${FLEET_SELF_MOUNT_PATH:-/opt/codex-fleet}
      - /docker:/docker
    extra_hosts:
      - "host.docker.internal:host-gateway"
    command: ["bash", "/opt/codex-fleet/scripts/run_chummer_design_supervisor.sh"]
    networks:
      - fleet_net
    healthcheck:
      test: ["CMD-SHELL", "python /opt/codex-fleet/scripts/healthcheck_design_supervisor.py"]
      interval: 30s
      timeout: 15s
      retries: 3
      start_period: 30s

  fleet-design-overwatch:
    build:
      context: ./controller
      args:
        CODEX_NPM_REFRESH_TOKEN: ${CODEX_NPM_REFRESH_TOKEN:-static}
    env_file:
      - ./runtime.ea.env
      - ./runtime.env
    container_name: fleet-design-overwatch
    restart: unless-stopped
    init: true
    user: "${LOCAL_UID:-1000}:${LOCAL_GID:-1000}"
    group_add:
      - "${DOCKER_GID:-112}"
    working_dir: ${FLEET_SELF_MOUNT_PATH:-/opt/codex-fleet}
    environment:
      OODA_WORKSPACE_ROOT: /docker/fleet
      OODA_SUPERVISOR_STATE_ROOT: /var/lib/codex-fleet/chummer_design_supervisor
      OODA_MONITOR_STATE_ROOT: /var/lib/codex-fleet/design_supervisor_ooda
      OODA_DURATION_SECONDS: "28800"
      OODA_POLL_SECONDS: "30"
      OODA_STALE_SECONDS: "180"
      OODA_CURRENT_ALIAS: current_8h
      OODA_FOREGROUND_MODE: "1"
      TZ: UTC
    volumes:
      - ./state:/var/lib/codex-fleet
      - ./config:/app/config:ro
      - ./secrets:/run/secrets:ro
      - ${HOME:-/home/tibor}/.codex/auth.json:/run/secrets/chatgpt.auth.json:ro
      - ${HOME:-/home/tibor}/.config/gh:/run/gh:ro
      - ./:${FLEET_SELF_MOUNT_PATH:-/opt/codex-fleet}
      - /docker:/docker
      - /var/run/docker.sock:/var/run/docker.sock
    extra_hosts:
      - "host.docker.internal:host-gateway"
    depends_on:
      - fleet-controller
      - fleet-design-supervisor
    command: ["bash", "/opt/codex-fleet/scripts/run_ooda_design_supervisor.sh"]
    networks:
      - fleet_net
    healthcheck:
      test:
        [
          "CMD-SHELL",
          "python - <<'PY'\nimport datetime as dt, json\nfrom pathlib import Path\npath = Path('/var/lib/codex-fleet/design_supervisor_ooda/current_8h/state.json')\nassert path.is_file(), path\npayload = json.loads(path.read_text(encoding='utf-8'))\nlast_cycle = str(payload.get('last_cycle_at') or '').strip()\nassert last_cycle, payload\nstamp = dt.datetime.fromisoformat(last_cycle.replace('Z', '+00:00'))\nassert (dt.datetime.now(dt.timezone.utc) - stamp).total_seconds() <= 900, last_cycle\nPY",
        ]
      interval: 30s
      timeout: 15s
      retries: 3
      start_period: 45s

  fleet-admin:
    build:
      context: ./admin
    env_file:
      - ./runtime.ea.env
      - ./runtime.env
    container_name: fleet-admin
    restart: unless-stopped
    init: true
    user: "${LOCAL_UID:-1000}:${LOCAL_GID:-1000}"
    environment:
      APP_PORT: 8092
      FLEET_CONFIG_PATH: /app/config/fleet.yaml
      FLEET_ACCOUNTS_PATH: /app/config/accounts.yaml
      FLEET_DB_PATH: /var/lib/codex-fleet/fleet.db
      FLEET_CODEX_HOME_ROOT: /var/lib/codex-fleet/codex-homes
      FLEET_GH_HOSTS_PATH: /run/gh/hosts.yml
      CHUMMER_HUB_REGISTRY_BASE_URL: ${CHUMMER_HUB_REGISTRY_BASE_URL:-http://host.docker.internal:18091}
      CHUMMER_RELEASE_REGISTRY_CURRENT_URL: ${CHUMMER_RELEASE_REGISTRY_CURRENT_URL:-}
      TZ: UTC
    volumes:
      - ./state:/var/lib/codex-fleet
      - ./config:/app/config
      - ./secrets:/run/secrets:ro
      - ${HOME:-/home/tibor}/.codex/auth.json:/run/secrets/chatgpt.auth.json:ro
      - ${HOME:-/home/tibor}/.config/gh:/run/gh:ro
      - ./:${FLEET_SELF_MOUNT_PATH:-/opt/codex-fleet}:ro
      - /docker:/docker
    extra_hosts:
      - "host.docker.internal:host-gateway"
    networks:
      - fleet_net
    healthcheck:
      test: ["CMD-SHELL", "python - <<'PY'\nimport urllib.request\nurllib.request.urlopen('http://127.0.0.1:8092/health', timeout=5)\nPY"]
      interval: 30s
      timeout: 5s
      retries: 3
      start_period: 20s

  fleet-auditor:
    build:
      context: ./auditor
    container_name: fleet-auditor
    restart: unless-stopped
    init: true
    user: "${LOCAL_UID:-1000}:${LOCAL_GID:-1000}"
    environment:
      APP_PORT: 8093
      FLEET_CONFIG_PATH: /app/config/fleet.yaml
      FLEET_DB_PATH: /var/lib/codex-fleet/fleet.db
      FLEET_AUDITOR_HEALTH_URL: http://127.0.0.1:8093/health
      FLEET_AUDITOR_RUN_MAX_AGE_SECONDS: "900"
      FLEET_AUDITOR_STARTUP_GRACE_SECONDS: "180"
      FLEET_STUDIO_SOURCE_PATH: /app/studio-src/app.py
      TZ: UTC
    volumes:
      - ./state:/var/lib/codex-fleet
      - ./config:/app/config:ro
      - ./studio:/app/studio-src:ro
      - ./:${FLEET_SELF_MOUNT_PATH:-/opt/codex-fleet}:ro
      - /docker:/docker:ro
    networks:
      - fleet_net
    healthcheck:
      test: ["CMD-SHELL", "python /opt/codex-fleet/scripts/healthcheck_auditor.py"]
      interval: 30s
      timeout: 15s
      retries: 3
      start_period: 30s

  fleet-quartermaster:
    build:
      context: ./quartermaster
    env_file:
      - ./runtime.ea.env
      - ./runtime.env
    container_name: fleet-quartermaster
    restart: unless-stopped
    init: true
    user: "${LOCAL_UID:-1000}:${LOCAL_GID:-1000}"
    environment:
      APP_PORT: 8094
      FLEET_CONFIG_PATH: /app/config/fleet.yaml
      FLEET_DB_PATH: /var/lib/codex-fleet/fleet.db
      FLEET_ADMIN_URL: http://fleet-admin:8092
      TZ: UTC
    volumes:
      - ./state:/var/lib/codex-fleet
      - ./config:/app/config:ro
      - ./:${FLEET_SELF_MOUNT_PATH:-/opt/codex-fleet}:ro
      - /docker:/docker:ro
    extra_hosts:
      - "host.docker.internal:host-gateway"
    networks:
      - fleet_net
    healthcheck:
      test: ["CMD-SHELL", "python - <<'PY'\nimport urllib.request\nurllib.request.urlopen('http://127.0.0.1:8094/health', timeout=5)\nPY"]
      interval: 30s
      timeout: 5s
      retries: 3
      start_period: 20s

  fleet-dashboard:
    image: nginx:1.27-alpine
    container_name: fleet-dashboard
    restart: unless-stopped
    depends_on:
      fleet-controller:
        condition: service_healthy
      fleet-studio:
        condition: service_healthy
      fleet-admin:
        condition: service_healthy
      fleet-auditor:
        condition: service_healthy
      fleet-quartermaster:
        condition: service_healthy
    volumes:
      - ./gateway/nginx.conf:/etc/nginx/nginx.conf:ro
      - ./gateway/static:/usr/share/nginx/html:ro
    ports:
      - 127.0.0.1:${HOST_DASHBOARD_PORT:-18090}:8090
    networks:
      - fleet_net
    healthcheck:
      test: ["CMD-SHELL", "wget -T 10 -q -O - http://127.0.0.1:8090/health >/dev/null 2>&1 && wget -T 10 -q -O - http://127.0.0.1:8090/ >/dev/null 2>&1"]
      interval: 30s
      timeout: 15s
      retries: 5
      start_period: 20s

  fleet-rebuilder:
    build:
      context: ./controller
      args:
        CODEX_NPM_REFRESH_TOKEN: ${CODEX_NPM_REFRESH_TOKEN:-static}
    container_name: fleet-rebuilder
    restart: unless-stopped
    init: true
    env_file:
      - ./runtime.ea.env
      - ./runtime.env
    environment:
      FLEET_REBUILD_COMPOSE_FILE: /docker/fleet/docker-compose.yml
      FLEET_REBUILD_STATE_DIR: /docker/fleet/state/rebuilder
      TZ: UTC
    volumes:
      - /var/run/docker.sock:/var/run/docker.sock
      - /docker:/docker
    command: ["sh", "/docker/fleet/scripts/rebuild-loop.sh"]
    networks:
      - fleet_net
    healthcheck:
      test: ["CMD-SHELL", "test -f /docker/fleet/state/rebuilder/heartbeat.txt"]
      interval: 30s
      timeout: 5s
      retries: 3
      start_period: 10s

networks:
  fleet_net:
    name: ${FLEET_NETWORK_NAME:-codex-fleet-net}
    driver: bridge