AI System Monitor - Resource Pressure Decoder.yaml

alias: AI System Monitor - Resource Pressure Decoder
description: >-
  When sustained memory pressure, IO pressure, memory use, or CPU use crosses a
  starter threshold, send a small System Monitor snapshot to an Ollama-backed AI
  Task and create one plain-language persistent notification. Tune the
  thresholds, CPU-core count, and entity IDs for your own Home Assistant host.
triggers:
  - trigger: numeric_state
    entity_id:
      - sensor.system_monitor_memory_pressure_full_60s_average
    above: 5
    for:
      hours: 0
      minutes: 1
      seconds: 0
    id: memory_pressure_full
  - trigger: numeric_state
    entity_id:
      - sensor.system_monitor_io_pressure_full_60s_average
    above: 5
    for:
      hours: 0
      minutes: 1
      seconds: 0
    id: io_pressure_full
  - trigger: numeric_state
    entity_id:
      - sensor.system_monitor_memory_usage
    above: 90
    for:
      hours: 0
      minutes: 1
      seconds: 0
    id: memory_high
  - trigger: numeric_state
    entity_id:
      - sensor.system_monitor_processor_use
    above: 90
    for:
      hours: 0
      minutes: 1
      seconds: 0
    id: cpu_high
conditions: []
actions:
  - variables:
      cpu_core_count: 4
      memory_entity: sensor.system_monitor_memory_usage
      cpu_entity: sensor.system_monitor_processor_use
      cpu_temperature: sensor.system_monitor_processor_temperature
      load_1_entity: sensor.system_monitor_load_1_min
      load_5_entity: sensor.system_monitor_load_5_min
      load_15_entity: sensor.system_monitor_load_15_min
      uptime_entity: sensor.system_monitor_uptime
      memory_pressure_some_entity: sensor.system_monitor_memory_pressure_some_60s_average
      memory_pressure_full_entity: sensor.system_monitor_memory_pressure_full_60s_average
      io_pressure_some_entity: sensor.system_monitor_io_pressure_some_60s_average
      io_pressure_full_entity: sensor.system_monitor_io_pressure_full_60s_average
      cpu_pressure_some_entity: sensor.system_monitor_cpu_pressure_some_60s_average
      trigger_reason: |-
        {{ trigger.id if trigger is defined and trigger.id is defined
           else 'manual_test' }}
      fallback_report: >-
        Home Assistant detected or manually tested a system-health condition,
        but the AI report was not available. Check History and Traces for
        processor use, memory use, load, and pressure. Do not restart services
        or delete anything until you know what caused the sustained reading.
  - alias: Create an immediate deterministic warning before calling AI
    action: persistent_notification.create
    data:
      notification_id: ai_system_monitor_resource_pressure
      title: |-
        {% set labels = {
          'memory_pressure_full': 'Memory pressure stayed high',
          'io_pressure_full': 'IO pressure stayed high',
          'memory_high': 'Memory use stayed high',
          'cpu_high': 'Processor use stayed high',
          'manual_test': 'Manual system-health test'
        } %} {{ labels.get(trigger_reason, 'System-health report') }}
      message: >-
        Home Assistant fired the deterministic {{ trigger_reason }} check. The
        AI Task is translating the current PSI, CPU, memory, load, and uptime
        values now. No restart, cleanup, settings change, or control action has
        run.
  - alias: Ask Ollama AI Task to translate the System Monitor snapshot
    continue_on_error: true
    action: ai_task.generate_data
    data:
      entity_id: ai_task.ollama_ai_task_gpt_oss_120b_cloud
      task_name: AI System Monitor Resource Pressure Report
      instructions: >-
        You are translating a small Home Assistant System Monitor snapshot into
        normal language. A deterministic Home Assistant trigger fired; you did
        not choose the trigger. The trigger ID is {{ trigger_reason }}.

        CPU cores configured by the user: {{ cpu_core_count }} Processor use: {{
        states(cpu_entity) }}% Memory use: {{ states(memory_entity) }}% Load
        average, 1 minute: {{ states(load_1_entity) }} Load average, 5 minutes:
        {{ states(load_5_entity) }} Load average, 15 minutes: {{
        states(load_15_entity) }} Last boot time: {{ states(uptime_entity) }}
        Memory pressure some, 60-second average: {{
        states(memory_pressure_some_entity) }}% Memory pressure full, 60-second
        average: {{ states(memory_pressure_full_entity) }}% IO pressure some,
        60-second average: {{ states(io_pressure_some_entity) }}% IO pressure
        full, 60-second average: {{ states(io_pressure_full_entity) }}% CPU
        pressure some, 60-second average: {{ states(cpu_pressure_some_entity)
        }}%

        Ignore unknown, unavailable, or missing readings. In no more than 140
        words, do all of the following: 1. Start with one plain verdict: normal,
        busy, or struggling. 2. Explain the trigger and the useful PSI readings
        in normal language. 3. Compare the 1, 5, and 15 minute load trend with
        the configured CPU-core
           count, but do not call a load value dangerous from one snapshot alone.
        4. Give three safe checks in order.

        Do not claim a root cause that these readings cannot prove. Do not tell
        the user to delete files, restart Home Assistant, stop add-ons, or
        change settings automatically. Do not use emojis.

        5. Include in the report the CPU temperature {{ states(cpu_temperature)
        }}% and if the levels are too high for a regular x86 or ARM cpu say
        that.

        6. Translate the last boot time into simple uptime context when
        available and list it only if it is above 20 days recommending to plan a
        restart just 

        7. Format the text to be easily readable (or playable from speakers)
        with bullets or something 
    response_variable: ai_response
  - alias: Create or update one plain-language system-health notification
    action: persistent_notification.create
    data:
      notification_id: ai_system_monitor_resource_pressure
      title: |-
        {% set labels = {
          'memory_pressure_full': 'Memory pressure stayed high',
          'io_pressure_full': 'IO pressure stayed high',
          'memory_high': 'Memory use stayed high',
          'cpu_high': 'Processor use stayed high',
          'manual_test': 'Manual system-health test'
        } %} {{ labels.get(trigger_reason, 'System-health report') }}
      message: |-
        {{ ai_response.data
           if ai_response is defined and ai_response.data is defined
           else fallback_report }}
mode: single
max_exceeded: silent