capability: document.ocr
version: 1
description: OCR and structured document extraction.
input_schema:
  type: object
  additionalProperties: false
  required: [content_base64, media_type]
  properties:
    content_base64: {type: string, minLength: 4, maxLength: 11184812}
    media_type: {enum: [image/png, image/jpeg]}
    language_hint: {enum: [nl, en, auto]}
output_schema:
  type: object
  additionalProperties: false
  required: [text, pages]
  properties:
    text: {type: string}
    pages: {type: array, minItems: 1, maxItems: 1}
    blocks: {type: array}
    confidence: {type: [number, 'null']}
modalities: {input: [image, document], output: [text, json]}
languages: [nl, en]
structured_output: true
quality_metrics: [cer, wer, field_accuracy, layout_accuracy]
upgrade_class: schema_breaking
fallback: {allowed: false, mode: hard_fail}
privacy: {classification: restricted, allow_persistence: false, allow_logging_payloads: false, allow_network_egress: false}
resources: {accelerator_required: true, cpu_fallback_allowed: false}
production_priority: background
default_residency: load_on_demand
estate:
  category: DOCUMENT
  purpose: Offline text extraction from bounded local document images.
  stability: experimental
  resource_class: MEDIUM
  evaluation_type: ocr
  consumers: [examplerag, examplevision, example-ops]
  payload_limits: {max_bytes: 8388608, max_batch_count: 1, max_width: 4096, max_height: 4096, max_pages: 1}