The ai_model_pipeline_steps core model exposes persisted AI pipeline step events from the observability store.

It is read-only and intended for monitoring/debug views that need to reconstruct how a model request moved through pipeline stages.

What This Model Represents¶

Rows include:

• id
• timestamp
• pipeline_id
• request_id
• step_id
• parent_step_id
• root_method
• type
• name
• status
• duration_ms
• provider
• engine
• model_registry_id
• model_name
• error

Detail rows also include serialized input_json, output_json, stats_json, and metadata_json.

Typical Use Cases¶

Use module_sdk.models.ai_model_pipeline_steps when you need to:

• show pipeline execution flows
• inspect failed or slow steps
• filter steps by pipeline, request, step name, provider, model, status, or model registry id
• build operational monitor pages without querying observability tables directly

Listing Pipeline Steps¶

listing = module_sdk.models.ai_model_pipeline_steps.list(
    page=0,
    page_size=50,
    filters={"pipeline_id": "pipeline_123"},
    sort={"field": "timestamp", "direction": "asc"},
)

Supported filters include:

• pipeline_id
• request_id
• type
• name
• status
• provider
• model_name
• model_registry_id

The default sort is timestamp asc, so steps render in execution order.

Table Schema¶

Use table_model() to build a DataTable aligned with the core schema:

model = module_sdk.models.ai_model_pipeline_steps.table_model()

The monitor module uses this schema for its pipeline flow table.

Write Operations Are Disabled¶

This model is read-only through the SDK. Pipeline step rows are generated by core AI pipeline instrumentation, not by module code.