Concept card

SAP Prompt Registry

As of 2026-07-24T14:00:00Z

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 Prompt Registry?

The Registry exists to close one specific compliance gap — 'shadow prompts' edited ad hoc with no record of which version drove a given agent decision.

The SAP Prompt Registry is the version-control and approval system for the single artifact that determines what a Joule agent actually does with the context it retrieves: the prompt. In this context a prompt is not a one-off question typed into a chat box — it is the system instruction, the user-prompt template, and any few-shot examples that together specify a Joule Skill's behavior. Because that instruction is the difference between an agent giving safe, on-brand, accurate answers and one that leaks information or behaves inconsistently, the Prompt Registry treats prompts the way mature software engineering treats code: versioned, reviewed, approved before deployment, and auditable after the fact.

The problem it exists to solve

Why it matters

Under EU AI Act Articles 9 and 13, deployers must produce evidence of which prompt version drove a given decision — shadow prompts make that impossible
Prompt changes require a designated reviewer (role AI_PROMPT_REVIEWER) to approve before deployment, not silent hotfixes
Every version (e.g. system-prompt@v3) stays retrievable for audit even after being superseded

Key points

Versioned prompt storage: every prompt is a named, versioned artefact — `prompts/<skill>/<type>@v3`; prior versions retained for audit.
Review-and-approval workflow: draft → AI Governance Architect review → approve/reject; only approved versions deployable to production.
Deployment gate: Joule Studio pipeline fails if any agent prompt reference is not in Registry at an approved version.
Audit logging: every inference call logs the prompt reference in the SAP Audit Journal — EU AI Act Art. 13 evidence trail at query time.
Shadow prompt problem: prompts in local files or env vars are not auditable — the Registry eliminates this compliance gap.
Model upgrade support: A/B test new prompt version in sandbox, approve winner, deploy with full audit trail showing version history.
Reviewer role: `AI_PROMPT_REVIEWER` SAP access control role — typically assigned to AI Governance Architect.
SAP Prompt Registry is mastered only when it changes a named buyer decision.
Start with the semantic contract and control model before demonstrating the tool.
Use current SAP, analyst, study, KG, and news signals as evidence, not decoration.

Terms used on this page

Shadow prompt: A prompt used in a production AI application that is not stored in a version-controlled, auditable repository — typically exists as a hardcoded string, an environment variable, or a local file; creates an EU AI Act Article 13 compliance gap.
AI_PROMPT_REVIEWER: SAP access control role that grants permission to review and approve prompt version changes in the SAP Prompt Registry; typically assigned to the AI Governance Architect on the programme.
Prompt versioning: The practice of assigning a version identifier (v1, v2, v3) to each variant of a prompt artefact, retaining all prior versions in an auditable store, and tracking which version was deployed to production at each point in time.
Deployment gate: An automated check in the CI/CD pipeline that verifies a precondition before allowing a deployment to proceed — in the SAP context, the Prompt Registry gate verifies that all prompt references in an agent workflow are at approved Registry versions.
Decision owner: The accountable person who accepts the trade-off and funds the next action.
Semantic contract: The shared definition of business terms, metrics, entities, and access rules used by tools and teams.
Control plane: The layer that applies policy, access, lineage, monitoring, and escalation across the operating model.
Evidence grade: A label that separates verified fact, directional signal, modeled assumption, and field observation.

Sources

Full card available to members. What the full card adds: the full decision framework · the SAP vs Snowflake / Databricks / Fabric comparison · the common pitfalls and their fix · the cheat sheet · the architecture schemas · the code blocks · the facts worth quoting.

Open in the app →