name: Argilla Vocabulary
description: Domain vocabulary for the Argilla data annotation and curation platform API
url: https://argilla.io
version: 2.8.0
created: 2026-06-12
modified: 2026-06-12

terms:

  - term: dataset
    label: Dataset
    definition: >
      A named collection of records to be annotated, associated with a workspace.
      Datasets define fields, questions, metadata properties, and vector settings
      that structure the annotation task.
    related:
      - record
      - workspace
      - field
      - question

  - term: record
    label: Record
    definition: >
      A single data item within a dataset that annotators review and label.
      Records contain field values (the data to annotate), metadata, responses
      (human annotations), suggestions (model predictions), and vector embeddings.
    related:
      - dataset
      - response
      - suggestion
      - field

  - term: field
    label: Field
    definition: >
      A typed input column within a dataset that holds the data to be annotated.
      Supported field types include text, image, chat, and custom fields.
    types:
      - text
      - image
      - chat
      - custom
    related:
      - dataset
      - record

  - term: question
    label: Question
    definition: >
      An annotation task presented to the annotator for each record. Questions
      define what annotators should label and support types such as label
      selection, multi-label selection, rating, ranking, span, and text.
    types:
      - label_selection
      - multi_label_selection
      - rating
      - ranking
      - span
      - text
    related:
      - dataset
      - response

  - term: response
    label: Response
    definition: >
      A human annotator's answer to one or more questions on a record. Responses
      have a status of submitted, discarded, or draft, and contain values keyed
      by question name.
    statuses:
      - submitted
      - discarded
      - draft
    related:
      - record
      - question
      - user

  - term: suggestion
    label: Suggestion
    definition: >
      An automated or model-generated pre-annotation for a record question.
      Suggestions can include a score and agent identifier, and serve as
      starting points for human annotators.
    related:
      - record
      - question

  - term: workspace
    label: Workspace
    definition: >
      A logical grouping namespace that contains datasets and controls user
      access. Users must be assigned to a workspace to view or annotate its
      datasets.
    related:
      - dataset
      - user

  - term: user
    label: User
    definition: >
      An account that can annotate records or administrate workspaces and
      datasets. Users authenticate via API key and have roles of owner or
      annotator.
    roles:
      - owner
      - annotator
    related:
      - workspace
      - response

  - term: metadata_property
    label: Metadata Property
    definition: >
      A typed descriptor attached to records for filtering and faceting.
      Supports integer, float, and terms (categorical) types, each with
      optional value ranges.
    types:
      - integer
      - float
      - terms
    related:
      - dataset
      - record

  - term: vector_settings
    label: Vector Settings
    definition: >
      Configuration for storing dense embedding vectors on records within a
      dataset. Defines the vector dimension and name, enabling similarity
      search across records.
    related:
      - dataset
      - record

  - term: webhook
    label: Webhook
    definition: >
      An HTTP callback registered to receive server-sent events when specific
      dataset or record lifecycle events occur (e.g. record created, response
      submitted).
    events:
      - dataset.created
      - dataset.updated
      - dataset.deleted
      - record.created
      - record.updated
      - record.deleted
      - response.created
      - response.updated
      - response.deleted
    related:
      - dataset
      - response

  - term: rlhf
    label: RLHF
    definition: >
      Reinforcement Learning from Human Feedback — a training technique in
      which human preference annotations collected through Argilla are used
      to fine-tune or align large language models.
    related:
      - response
      - suggestion
      - dataset

  - term: hub_dataset
    label: Hub Dataset
    definition: >
      A Hugging Face Hub dataset reference used to import records into Argilla
      or export annotated records back to the Hub for model training.
    related:
      - dataset
      - record