capability: vision.embedding
version: 1
description: Image/text embeddings for visual retrieval.
input_schema:
  type: object
  additionalProperties: false
  required: [items]
  properties:
    items:
      type: array
      minItems: 1
      maxItems: 4
      items:
        oneOf:
          - {type: object, additionalProperties: false, required: [image_base64, media_type], properties: {image_base64: {type: string}, media_type: {enum: [image/png, image/jpeg]}}}
          - {type: object, additionalProperties: false, required: [text], properties: {text: {type: string, minLength: 1, maxLength: 4096}}}
output_schema:
  type: object
  required: [vectors]
  properties:
    vectors: {type: array, items: {type: array, items: {type: number}}}
modalities: {input: [image, text], output: [vector]}
languages: [nl, en]
vector: {dimensionality: null, normalized: null, cross_deployment_compatible: false}
quality_metrics: [recall_at_1, recall_at_5, recall_at_10, mrr]
upgrade_class: requires_reindex
fallback: {allowed: false, mode: hard_fail}
privacy: {classification: confidential, allow_persistence: false, allow_logging_payloads: false, allow_network_egress: false}
resources: {accelerator_required: true, cpu_fallback_allowed: false}
production_priority: interactive
default_residency: load_on_demand
estate:
  category: VISION
  purpose: Shared image and text vector space for local visual retrieval.
  stability: experimental
  resource_class: HEAVY
  evaluation_type: visual-retrieval
  consumers: [examplevision]
  payload_limits: {max_bytes: 8388608, max_batch_count: 4, max_width: 4096, max_height: 4096}