SAP Generative AI Hub — Hands-on: Orchestration, Grounding, Masking, Filtering
As of 2026-09-23
Hands-on lab for the generative AI hub in SAP AI Core (part of AI Foundation in the SAP Business AI Platform). You provision the extended plan (a one-way door), reuse or create the orchestration deployment, call POST /v2/completion from the SAP Cloud SDK for AI, and switch on templating, grounding (help.sap.com, Vector API, managed pipelines), data masking (SAP Data Privacy Integration) and content filtering (Azure Content Safety, Llama Guard 3). Prompts move into the prompt registry with Git-synced declarative templates. Two dated actions: orchestration v1 is decommissioned on 31 October 2026, and masking_providers gave way to providers. Costs are tokens → GenAI tokens → capacity units per SAP Note 3437766 — never a made-up price.
What you will learn
- Provision the generative AI hub correctly: SAP AI Core extended plan, SAP AI Launchpad standard plan, service key, workspace connection and the four genai/prompt role collections — and explain why the extended plan cannot be downgraded
- Locate or create an orchestration deployment (scenario and executable 'orchestration') in a named resource group and call POST /v2/completion from the SAP Cloud SDK for AI in TypeScript and Python
- Configure the five orchestration v2 modules — prompt_templating, grounding, masking, filtering, translation — and read intermediate_results to prove what the model actually received
- Ground answers on help.sap.com, on a Vector API collection and on a managed pipeline (SharePoint, S3, SFTP, Work Zone, Document Management, ServiceNow, Google Drive), respecting the document-grounding label and pipeline limits
- Version prompts and orchestration configs in the prompt registry (imperative for design time, declarative Git sync for runtime) and explain the token → GenAI token → capacity unit billing chain without inventing prices
Module overview
This is a lab module, not a brochure. By the end you will have provisioned the generative AI hub in SAP AI Core, located (or created) an orchestration deployment, called it from the SAP Cloud SDK for AI in TypeScript and Python, and switched on the four modules that make an LLM call defensible in an enterprise: templating, grounding, data masking and content filtering — plus the prompt registry that turns prompts into versioned, reviewable assets. Every step below is checkable: each exercise ends with an observable result you can show to a reviewer.
Naming, as of September 2026. The generative AI hub is a capability of SAP AI Core, which SAP now presents as part of "AI Foundation" inside the SAP Business AI Platform announced at Sapphire 2026 (BTP + Business Data Cloud + AI Foundation). The product you provision in the BTP cockpit is still called SAP AI Core; the cockpit UI is SAP AI Launchpad. There is no separate "SAP AI Hub" product to buy — when a client says "AI Hub", map it to the generative AI hub in SAP AI Core.
Prerequisites
- Access to an SAP BTP global account where you can assign the SAP AI Core extended plan (or the 30-day generative AI hub trial)
- Basic Node.js/TypeScript or Python, and comfort with curl and JSON
- Review core concepts first: C119, C026, C027
Outcomes
- Stand up a working generative AI hub tenant and prove it with a RUNNING orchestration deployment and a first templated completion.
- Build a grounded, masked, filtered orchestration call and explain each intermediate result to a security reviewer.
- Move a prompt from Launchpad experimentation to a Git-managed, versioned template referenced by scenario, name and version.
- Estimate and monitor consumption with SAP Note 3437766, the AI Core cost calculator and the BTP Usage export — without quoting invented prices.
Full module available to members. The full module adds: the decision framework · the end-to-end scenario walkthrough · the KPI scorecard · the anti-patterns · the code blocks · the knowledge check · the diagrams.