capability: rag.embedding
version: 1
description: Multilingual text embeddings for semantic retrieval.
input_schema:
  type: object
  additionalProperties: false
  required: [input]
  properties:
    input:
      oneOf:
        - {type: string, minLength: 1, maxLength: 8192}
        - {type: array, items: {type: string, minLength: 1, maxLength: 8192}, minItems: 1, maxItems: 8}
output_schema:
  type: object
  additionalProperties: false
  required: [vectors, dimension, normalized, embedding_space_id]
  properties:
    vectors: {type: array, minItems: 1, maxItems: 8, items: {type: array, minItems: 1024, maxItems: 1024, items: {type: number}}}
    dimension: {const: 1024}
    normalized: {const: true}
    embedding_space_id: {type: string, format: uuid}
modalities: {input: [text], output: [vector]}
languages: [nl, en]
vector: {dimensionality: 1024, normalized: true, cross_deployment_compatible: false}
slo: {latency_p95_ms: 15000, availability_percent: 99.0, max_concurrency: 1}
quality_metrics: [recall_at_5, recall_at_10, mrr, ndcg_at_10]
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: production
default_residency: keep_warm
estate:
  category: RAG
  purpose: Multilingual semantic retrieval for local knowledge projects.
  stability: stable
  resource_class: MEDIUM
  evaluation_type: retrieval
  consumers: [examplerag, portfolio-assistant]
  payload_limits: {max_bytes: 65536, max_batch_count: 8}