Actions¶

Five actions answer inside an automation and return a response variable.

- action: jev.noul
  response_variable: laundry
  target:
    entity_id: sensor.washing_machine_power
  data:
    instructions: Is the laundry finished but still sitting in the machine?
    background: >-
      This machine draws under 5 W when idle and over 300 W while a programme runs.
    threshold: 0.7
- if: "{{ laundry.is_true }}"
  then:
    - action: notify.mobile_app
      data: { message: The washing is done and still in the machine. }

Action	You give it	You get back
`jev.noul`	a yes/no question	`noul` 0 to 1, `is_true` against your threshold
`jev.choice`	`options:`, 2 to 255	`choice`, `probabilities`, `confidence`
`jev.score`	`levels:`, 2 to 10, lowest first	`score`, `normalized`, `nearest_level`, `legend`, `probabilities`, `confidence`
`jev.ask`	any mix, under your own keys	the same, under `answers`
`jev.box`	one camera or image entity, and what to find in it	`left`, `top`, `right`, `bottom`, `confidence`, `width`, `height`

Every response also carries usage and latency_ms, so an automation can watch what it spends.

Point at entities instead of writing a template¶

All four take the normal Home Assistant target picker. Select entities, devices, areas, floors or labels and the state is built for you:

- action: jev.choice
  response_variable: room
  target:
    area_id: kitchen
  data:
    instructions: What is happening in this room?
    options:
      cooking: Somebody is preparing food
      eating: Somebody is at the table
      empty: Nobody is in here

Up to 250 entities. Past that it refuses and tells you your number, because every entity is billed on every evaluation.

The actions send whatever you target, whether or not it is exposed to Assist. An automation names its entities on purpose, so the Assist exposure list is not consulted here; it is consulted for the conversation agent. include_attributes: true never sends access tokens, entity pictures or coordinates, so a camera or a device_tracker is safe to target.

Show it a camera¶

images: takes camera and image entities, and sends the current picture of each beside the state:

- action: jev.noul
  response_variable: parcel
  data:
    images: [camera.front_door]
    instructions: Is there a parcel on the doorstep?

It works alone or with a target and a state. A picture costs hundreds to thousands of tokens, and the budget counts it. Cameras and images has the sizes, the cost, jev.box and what each error means.

Ask another model for one call¶

Every action takes model:, which asks that model instead of the entry's for this call only. The entry's model, its version sensor and its options stay as they are. Box questions need it today: a new entry asks jev-latest, which answered as Jev 1.13 on 2026-10-16, and Box needs jev-1.14.0 or jev-preview.

More than one entry¶

Each Jev entry has its own key and budget. With more than one loaded, an action has to name the one to ask with under API key (config_entry: in YAML). An action that names none is refused before anything is sent, and the error lists the entries.

Ask several things at once¶

jev.ask takes any mix under your own keys, in one request:

- action: jev.ask
  response_variable: jev
  data:
    state:
      transcript: "{{ trigger.payload }}"
    questions:
      caller:
        type: choice
        instructions: Who is at the door?
        criteria:
          delivery: A courier
          visitor: Someone we know
      urgency:
        type: score
        instructions: How quickly does this need a person?
        criteria: [Never, Whenever, Right now]
      answer_now:
        type: noul
        instructions: Does someone need to go to the door right now?

This is the shape to reach for. Adding a question costs tokens, not time, so asking the five you might need and discarding four is cheaper than two round trips.

Templates and variables¶

state: takes a template, an object or a list, and automation variables reach it:

- variables:
    household: "{{ states('input_text.house_notes') }}"
- action: jev.noul
  response_variable: answer
  data:
    state:
      note: "{{ household }}"
      time: "{{ now().strftime('%H:%M on %A') }}"
    instructions: Is this worth waking someone for?

An object arrives as an object, not as a stringified dict. The model reads the field names as labels, so name them for what they hold.

Tune a threshold¶

jev.calibrate is a fifth action that asks nothing. It compares the recorded history of a noul sensor with an entity that shows what was really true, and returns the threshold that fits best. It costs no tokens. See tuning a threshold.

Errors name the limit¶

Every budget failure names the budget, the limit and your ask, because an agent reading the error can fix max 255 options, asked for 300 and cannot fix a blank window.