Concept card

SAP AI Core resource groups, scenarios and executables

As of 2026-09-25

Edited by Cédric-Olivier Mâry, sole editor and publication director — 27 years in the SAP ecosystem. How these pages are produced, and where automation is used.

What is SAP AI Core resource groups, scenarios and executables?

SAP AI Core's object model has six parts: scenarios and executables (templates), configurations (concrete parameter instances), executions and deployments (runs), and artifacts (data flowing between them). Resource groups isolate executions, deployments, configurations and artifacts per team or workload — but scenarios, executables and Docker registry secrets are shared tenant-wide, with a documented ceiling of 50 resource groups per tenant.

The vocabulary, precisely

SAP AI Core's own documentation defines six objects, and getting their relationships right is the difference between reading an AI Core project and merely poking at its UI. A scenario is an implementation of a specific AI use case: a predefined set of AI capabilities in the form of executables and templates, grouping related work and tracking versions as models evolve — SAP's own example scenario ID is "foundation-models." An executable is a reusable template defining a workflow or pipeline — training a model, running inference, creating a deployment — and comes in two kinds: non-deployable executables, which produce executions, and deployable executables, which produce deployments. A configuration is the concrete instance of an executable: a collection of parameters, artifact references (datasets, models) and environment settings that instantiate and run an execution or deployment; one executable can have many configurations. An execution is a single run — training executions produce trained models, batch-inferencing executions produce result sets. A deployment is an instance of a model-serving template configured with a model artifact and applied to data in a serving request; it creates a model server and an inference URL. And an artifact is any tangible by-product an execution or deployment consumes or produces — data, files, binaries, libraries — flowing between the others as input or output through the AI lifecycle.

Why it matters

Scenarios and executables are shared tenant-wide while executions, deployments, configurations and artifacts are isolated per resource group — misreading which objects are shared versus isolated leads to broken multi-team access assumptions.
SAP's own security guidance explicitly warns against reusing the same object store bucket and AWS IAM credentials across resource groups, because that silently defeats the isolation the resource group boundary is meant to provide.
Deleting a resource group deprovisions any scenario-consumer tenants inside it outright under the XSUAA multitenancy model — not a reversible cleanup action, worth flagging before anyone runs it in a shared tenant.

Key points

Scenario: a predefined set of executables and templates implementing a specific AI use case, with version tracking (example ID: 'foundation-models').
Executable: a reusable workflow/pipeline template; non-deployable executables produce executions, deployable executables produce deployments.
Configuration: the concrete parameter/artifact-reference/environment instance of an executable, created via POST /v2/lm/configurations naming a scenarioId and executableId; one executable can have many configurations.
Execution vs Deployment: an execution is a single run (training → trained model; batch inference → result set); a deployment is a standing model server with an inference URL.
Artifact: any tangible data/file/binary/library by-product consumed or produced by an execution or deployment.
Resource groups: tenant-scoped isolation units, one default per tenant, max 50 per tenant, created/deleted via the AI API.
Shared tenant-wide: scenarios, executables, Docker registry secrets, workflow/serving templates, UAA. Isolated per resource group: executions, deployments, configurations, artifacts.
Service plan (Standard vs Extended) governs whether generative-AI scenarios are usable at all, independent of the resource-group/scenario/executable model.

Terms used on this page

Scenario: A predefined set of AI capabilities (executables and templates) implementing a specific AI use case, shared across all resource groups in a tenant.
Executable: A reusable workflow/pipeline template; non-deployable executables produce executions, deployable executables produce deployments; shared tenant-wide.
Configuration: The concrete parameter, artifact-reference and environment-setting instance of an executable, used to start an execution or deployment; isolated per resource group.
Execution: A single run of a non-deployable executable — training (produces a trained model) or batch inference (produces a result set); isolated per resource group.
Deployment: A running instance of a model-serving template configured with a model artifact, exposing an inference URL; isolated per resource group.
Artifact: Any tangible by-product (data, file, binary, library) consumed or produced by an execution or deployment; isolated per resource group.
Resource group: A tenant-scoped isolation namespace for executions, deployments, configurations and artifacts; up to 50 per tenant, one default created automatically.

Sources

Full card available to members. What the full card adds: the full decision framework · the common pitfalls and their fix · the cheat sheet · the code blocks · the facts worth quoting.

Open in the app →