create_variable declares a named value with a type; set_variable updates one later in the pipeline. There are four types: string, number, boolean (collectively "scalar"), entity, and collection.

Scalar (string / number / boolean)

A single typed value. If you don't give an initial value, it defaults to that type's zero-value (empty string, 0, or false). Reference it downstream as {{myVar}}.

Entity

Holds one record — or null. The initial value must be an object (typically a {{step}} reference to a get_entity/create_entity output, or a literal record); anything else fails the step with a 400. Reference fields the same way you would off any entity step: {{myVar.fieldName}}.

Collection

Holds a list. Internally it's { items: [...], count, entityName?, scalarType? } — the same shape list_entities and an ai_call with outputFormat: list produce, which is why formulas and loops can treat all three interchangeably.

  • Entity collection — entityName is set, items are records of that entity.
  • Scalar collection — scalarType is set (string/number/boolean), items are plain values.
  • Shorthand accessors on any collection — {{myList.count}}, {{myList.first}}, {{myList.last}}.
  • Numeric aggregates — {{myList.min}}, {{myList.max}}, {{myList.avg}}, {{myList.sum}} — only meaningful (and only shown in the picker) for a scalar-number collection.

These accessors are {{...}}-only — not inside a formula

{{myList.sum}} etc. are computed by the same {{...}} resolver every non-formula step config uses. Reference them directly in another step's config (a create_entity field, a connector field mapping, a webhook body). A formula's own bare refs use a different, simpler resolver that doesn't compute these — and formula's COLLECTION_AVG/MAX/MIN/FIND functions only apply to entity collections (they look up a named field on each item), not scalar collections. So a scalar-number collection has no in-formula way to aggregate it — use the {{...}} accessor from a non-formula step instead.

set_variable — updating a variable later

Targets an existing variable by name with an operation:

  • `replace` — overwrite the whole value (any type).
  • `add` — type-dependent: sums numbers, concatenates strings, appends an item to a collection's items. Not valid for entity or boolean.
  • `clear` — resets to empty: []/0 items for a collection, null for an entity.

Worked example — build a collection in a loop, then aggregate

text 
s1: list_entities     entity=order_line  filters=[{field: order_id, op: eq, value: {{input.order_id}}}]
s2: create_variable    name=amounts  varType=collection  scalarType=number
s3: loop               source={{s1}}  itemAs=row
      s3a: set_variable  target=amounts  operation=add  value={{row.amount}}
s4: create_entity      entity=order_summary
                       fields=[{ order_id: {{input.order_id}} }, { total: {{amounts.sum}} }, { line_count: {{amounts.count}} }]

{{amounts.sum}}/{{amounts.count}} in s4 resolve correctly because they're template refs on a non-formula step's config — the same accessor would not resolve inside a formula's own expression syntax (see the warning above).

Formulas

The function library that reads collections and scalars.

Step types

Where create_variable and set_variable fit among every step.