capability: assistant.vision
version: 1
description: Visual question answering and structured image reasoning.
input_schema: {type: object, required: [image, prompt], properties: {image: {type: string}, prompt: {type: string}}}
output_schema: {type: object, required: [content], properties: {content: {type: string}, structured_output: {type: object}}}
modalities: {input: [image, text], output: [text, json]}
languages: [nl, en]
streaming: false
structured_output: true
quality_metrics: [task_success, structured_output_validity]
upgrade_class: behavioral
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: ASSISTANTS
  purpose: Future local image reasoning contract; not an M9 executable baseline.
  stability: planned
  resource_class: EXCLUSIVE_GPU
  evaluation_type: llm
  consumers: [examplevision, example-ops]
  payload_limits: {max_bytes: 8388608, max_batch_count: 1, max_width: 4096, max_height: 4096}