generated: '2026-06-20'
method: derived
source: openapi/tensorflow-serving-openapi.yml, https://www.tensorflow.org/tfx/serving/api_rest
authentication:
  style: none-builtin
  detail: See authentication/tensorflow-authentication.yml — access control is delegated to a fronting proxy/gateway/mesh.
content_type:
  request: application/json
  response: application/json
  note: All Serving REST inference requests and responses are JSON.
versioning:
  style: uri-path
  api: /v1
  model_selection:
    by_version: /v1/models/{model_name}/versions/{version}
    by_label: /v1/models/{model_name}/labels/{label}
  detail: See lifecycle/tensorflow-lifecycle.yml.
inference_formats:
  row:
    field: instances
    note: Array of individual inputs; response returns "predictions".
  column:
    field: inputs
    note: Map of tensor name to value arrays; response returns "outputs".
  rule: A predict request must set exactly one of "instances" or "inputs", never both.
  signature_selection:
    field: signature_name
    note: Optional; selects a named SignatureDef from the SavedModel.
pagination:
  supported: false
  note: Inference and model-status APIs are single-resource; no collection pagination.
idempotency:
  header: null
  supported: false
  note: No Idempotency-Key mechanism; inference calls are naturally repeatable but not deduplicated by the server.
request_tracing:
  request_id_header: null
  note: No provider request-id header; tracing is added by the operator's proxy/mesh.
rate_limiting:
  signaling: none-builtin
  note: The ModelServer does not emit rate-limit headers; throttling is an operator concern. See rate-limits/tensorflow-rate-limits.yml.
error_envelope:
  media_type: application/json
  field: error
  detail: See errors/tensorflow-problem-types.yml.
transports:
  - rest
  - grpc
grpc_note: The same models are served over gRPC (PredictionService/ModelService); see grpc/*.proto.