How to collect lists, ranges, preferences, and attachments

Back to how-to guides Use this pattern when one value is not enough: a list of items, a low and high range, or a file the customer uploads.

Concept

A GATHER field holds one scalar by default. Three properties change its shape. A separate construct handles files.

You need	Use	Stored shape
Several items	`list: true`	An array
A low and a high	`range: true`	An object with `low` and `high`
A choice from a fixed set	`options`, or `entity_ref`	One of the values
Preferences	See Preferences	Not a gather shape
A file or image	`AWAIT_ATTACHMENT` in a `FLOW` step	The attachment ID, as text

Use only one shape property on a field. A field with both list and range compiles without a warning, and the list wins. Write list: true and range: true in lowercase. yes and True read as false without an error.

Minimal working example

A hotel search collects a nightly budget range, a list of amenities, and a room type. The room type uses a reusable entity. The search tool is a project-local assumption. Bind it to your hotel search service in Project Tools. agents/hotel-search.agent.abl

AGENT: HotelSearch
GOAL: "Collect hotel search criteria: nightly budget range, amenities, and room type."

ENTITIES:
  room_type:
    TYPE: enum
    VALUES: [single, double, suite]
    SYNONYMS:
      single: [solo, one bed]
      double: [two beds, twin]

TOOLS:
  search_hotels(budget_low: number = null, budget_high: number = null, amenities: string = null, room_type: string = null) -> {results: array}
    description: "Searches hotels that match the collected criteria."
    confirm: never

GATHER:
  budget:
    type: number
    prompt: "What is your nightly budget range?"
    range: true
    semantics:
      unit: USD
  amenities:
    type: string
    prompt: "Which amenities do you need? Name as many as you like."
    list: true
    required: false
  room:
    entity_ref: room_type
    prompt: "Which room type do you want?"
  rating:
    type: enum
    options: [any, three_star, four_star, five_star]
    prompt: "What minimum star rating do you want?"
    required: false

COMPLETE:
  - WHEN: budget.low IS SET AND budget.high IS SET AND room IS SET
    RESPOND: "Searching {{room}} rooms from {{budget.low}} to {{budget.high}} USD per night."

Variable sources:

budget, amenities, room, and rating are GATHER fields. The customer supplies them.
COMPLETE reads budget.low, budget.high, and room.

How it works

How each shape is stored and checked:

A list is an array. A single value is wrapped into a one-item array. A text value is split on commas, so “pool, gym” gives two items. A semicolon does not split.
A range is an object with low, high, or both. At least one end is needed. Each end follows the field type, so a currency range of “$100 to $300” gives 100 and 300. The runtime accepts low greater than high.
An enum field accepts a value that matches an option exactly, then ignoring case, then through a synonym. It does not match approximately.
An entity_ref field inherits the entity type, values, synonyms, and sensitive setting. A field that sets entity_ref and also type or options is a compile error GATHER_ENTITY_REF_CONFLICT. An unknown entity is a compile error GATHER_ENTITY_REF_UNRESOLVED.

How to read the value:

In a template or condition	Result
`{{budget.low}}`	`100`
`{{amenities}}`	The array as multi-line JSON, not a comma list.
`{{amenities[0]}}` or `{{amenities.length}}`	The first item, or the count.
`budget.low IS SET AND budget.high IS SET`	Tests both ends. `budget IS SET` is true when only one end is set.
`"pool" IN amenities`	Tests list membership.
`amenities CONTAINS "pool"`	Always false. `CONTAINS` works on text only.
`amenities IS SET`	True for an empty list. Test `amenities.length > 0`.

A {{#each}} block does not iterate a list of strings. It works for a list of objects only. There is no join filter. How the semantics block helps:

temporal: future or past limits a date field to one side of today.
lookup checks the value against a lookup table. A near match becomes a confirmation.
unit with convert_to converts a number inside a FLOW step, and keeps the original. It does not convert in a top-level GATHER, and it skips lists and ranges.
format and components guide extraction only. They do not create a structured object.
locale and kore_entity_type carry metadata and have no effect on collection.

Do not put a validate rule on a list or range field. The rule applies to the whole value. A list with a pattern, and a range with a range rule, reject valid answers. The same applies to a multi-select enum: an enum rule from options accepts one value, so a list of two options fails.

Common variations

Range of dates and a unit conversion in a FLOW step

A FLOW step can convert units and collect a date range. agents/shipment-quote.agent.abl

AGENT: ShipmentQuote
GOAL: "Collect parcel weight, package types, and a travel window to quote a shipment."

TOOLS:
  quote_shipment(weight_lbs: number = null, package_types: string = null, window_start: string = null, window_end: string = null) -> {price: number}
    description: "Returns a shipping price for the collected parcel details."
    confirm: never

FLOW:
  collect -> quote -> done

  collect:
    REASONING: false
    GATHER:
      - weight:
          type: number
          prompt: "How much does the parcel weigh?"
          required: true
          semantics:
            unit: kg
            convert_to: lbs
      - package_types:
          type: enum
          options: [box, envelope, tube]
          prompt: "Which package type do you use?"
          required: true
      - ship_window:
          type: date
          prompt: "Between which dates can the parcel ship?"
          required: true
          range: true
          semantics:
            temporal: future
    THEN: quote

  quote:
    REASONING: false
    CALL: quote_shipment
      WITH:
        weight_lbs: "{{weight}}"
        package_types: "{{package_types}}"
        window_start: "{{ship_window.low}}"
        window_end: "{{ship_window.high}}"
      AS: quote_result
    THEN: done

  done:
    REASONING: false
    RESPOND: "Your quote is {{quote_result.price}}."

The converter supports temperature, distance, weight, time, volume, and currency units. A pair it does not support is skipped without an error.

Enum values

Write enum options in the block form or without quotes. A one-line list with quotes keeps the quote marks in the stored values. validate: enum(a, b) splits on commas. A pipe makes one value.

Preferences

preferences: true compiles, but it does not store a category object in the field. The field does not hold accept, desire, avoid, and refuse lists. A reasoning agent also scans each customer message for stated preferences. It saves them to memory under preferences.accept, preferences.desire, preferences.avoid, and preferences.refuse when the agent has a memory store. That scan runs on top-level GATHER fields only. To keep preferences in the conversation, collect them as a list: true field for each category, or use MEMORY and REMEMBER.

Attachments in a FLOW step

Use AWAIT_ATTACHMENT to wait for an upload. The step stores the first attachment ID in the named variable. A tool parameter named attachment_id is bound and checked against the available attachments. agents/id-upload.agent.abl

AGENT: IdUpload
GOAL: "Collect a photo of the customer's ID and submit it for review."

TOOLS:
  submit_id_photo(attachment_id: string = null) -> {status: string}
    description: "Submits the uploaded ID photo for review."
    confirm: never

MEMORY:
  session:
    - id_photo
      TYPE: string
      DESCRIPTION: "ID of the uploaded ID photo."

FLOW:
  start -> ask_photo -> submit -> done

  start:
    REASONING: false
    SET: id_photo = ""
    THEN: ask_photo

  ask_photo:
    REASONING: false
    AWAIT_ATTACHMENT:
      name: id_photo
      prompt: "Please upload a photo of your ID."
      category: image
      required: true
      timeout: 300
      on_timeout: skipped
    THEN: submit

  submit:
    REASONING: false
    CALL: submit_id_photo
      WITH:
        attachment_id: "{{id_photo}}"
      AS: submit_result
    THEN: done

  done:
    REASONING: false
    RESPOND: "Thanks. I received your ID photo."

  skipped:
    REASONING: false
    RESPOND: "I did not receive a photo. You can start again at any time."

The SET in the start step gives the compiler a producer for id_photo. Without it, the compiler warns that nothing produces the variable. Rules for AWAIT_ATTACHMENT:

category is image, document, audio, or video. A wrong category is a compile error INVALID_AWAIT_ATTACHMENT.
A file of the wrong category makes the step ask again. The trace outcome is category_mismatch.
With required: false, a text message sets the variable to null and the flow moves on.
With timeout in seconds and on_timeout, the flow moves to that step. Without on_timeout, the step replies that the upload timed out and keeps waiting.

The ATTACHMENTS section

The ATTACHMENTS: section declares upload fields with a category, a size limit, allowed types, and processing flags. The keys are prompt, category, required, max_size_mb, allowed_types, ocr_enabled, transcription_enabled, and key_frame_extraction. Keys such as allowed_mime_types and ocr are ignored without an error. A category outside the four valid values is also accepted silently. The section is a declaration only. The runtime does not read its size limit, allowed types, or processing flags. Uploads are limited by the project and tenant attachment settings: whether attachments are enabled, the maximum file size, the maximum files per session, the allowed MIME types, the PII policy, and the processing mode. Set these in the project settings for attachments. Use AWAIT_ATTACHMENT with a category to check the kind of file inside a flow.

Verification

Run the validator on each file. Expect no errors and no warnings.
Say “Between 100 and 300 dollars, with a pool and a gym, in a double”. Expect budget.low 100, budget.high 300, two amenities, and room double.
Say “around 200”. Expect COMPLETE not to fire until both ends are set.
In the shipment example, say “5 kg”. Expect the converted weight in pounds in the tool call.
In the upload example, send a document when an image is required. Expect the step to ask again.
Let the upload time out. Expect the skipped reply.
In the trace, open the dsl_collect events for the extracted values. Open the await_attachment events for the upload outcomes.

Production readiness checklist

Each field uses one shape property.
No list or range field has a validate rule.
Conditions test both ends of a range, and test .length for a list.
Templates do not print a raw list. They print items, or the count.
Enum options are in the block form, or unquoted.
Each upload step has a timeout and an on_timeout route.
Upload type, size, and count limits are set in the project attachment settings and tested.
Tests cover a one-item list, an empty list, a one-ended range, and a reversed range.

Common mistakes

Mistake	Why it happens	How to avoid it
Writing `list: yes`	Other languages accept it.	Write lowercase `true`.
Combining `list` and `range`	Both seem to apply.	Use one.
Adding `validate` to a list field	Validation looks harmless.	Remove it. The rule applies to the whole list.
Using `CONTAINS` on a list	It reads naturally.	Write `"item" IN list`.
Reading `range IS SET`	It looks like a full check.	Test `low` and `high`.
Expecting `preferences: true` to fill four lists	The documentation of the property suggests it.	Use list fields or `REMEMBER`.
Expecting unit conversion in a top-level `GATHER`	The property parses there.	Collect in a `FLOW` step.
Writing `allowed_mime_types` or `ocr`	Names look obvious.	Use `allowed_types` and `ocr_enabled`.

Troubleshooting

Symptom	Likely cause	What to check
A valid list is rejected	A `validate` rule or an enum rule applies to the whole list.	Remove `validate`. For a multi-select enum, collect as `string` with `list: true`.
A range stays unset	The customer gave one end, or the field also sets `list`.	The stored object and the shape properties.
A list prints as JSON	The template prints the whole array.	Print an item, or `.length`.
`{{#each}}` prints nothing	The list holds strings.	Use an index, or a list of objects.
An enum value is rejected	The value is not an exact or synonym match.	`options`, `SYNONYMS`, and the customer wording.
The compile fails with `GATHER_ENTITY_REF_CONFLICT`	`entity_ref` is set with `type` or `options`.	Remove `type` and `options`.
The compile warns that `id_photo` has no producer	Nothing sets the attachment variable.	Declare it in `MEMORY` and add a `SET` in an earlier step.

Related articles:

Get started

Build

Test and evaluate

Deploy

Analytics and insights

References

How to collect lists, ranges, preferences, and attachments

Concept

Minimal working example

How it works

Common variations

Range of dates and a unit conversion in a FLOW step

Enum values

Preferences

Attachments in a FLOW step

The ATTACHMENTS section

Verification

Production readiness checklist

Common mistakes

Troubleshooting

​Concept

​Minimal working example

​How it works

​Common variations

​Range of dates and a unit conversion in a FLOW step

​Enum values

​Preferences

​Attachments in a FLOW step

​The ATTACHMENTS section

​Verification

​Production readiness checklist

​Common mistakes

​Troubleshooting

Concept

Minimal working example

How it works

Common variations

Range of dates and a unit conversion in a FLOW step

Enum values

Preferences

Attachments in a FLOW step

The ATTACHMENTS section

Verification

Production readiness checklist

Common mistakes

Troubleshooting