Sap Ai Core

SAP AI Core & AI Launchpad Skill

Related Skills

• sap-btp-cloud-platform: Use for platform context, BTP account setup, and service integration
• sap-cap-capire: Use for building AI-powered applications with CAP or integrating AI services
• sap-cloud-sdk-ai: Use for SDK integration, AI service calls, and Java/JavaScript implementations
• sap-btp-best-practices: Use for production deployment patterns and AI governance guidelines

When to Use This Skill

Use this skill when provisioning SAP AI Core, using SAP AI Launchpad, configuring Generative AI Hub orchestration, choosing model providers, building RAG or grounding flows, managing prompt templates, deploying training/inference workloads, or wiring AI capabilities into SAP applications.

Table of Contents

1. Overview 2. Quick Start 3. Service Plans 4. Model Providers 5. Orchestration 6. Content Filtering 7. Data Masking 8. Grounding (RAG) 9. Tool Calling 10. Structured Output 11. Embeddings 12. ML Training 13. Deployments 14. Bundled Resources 15. SAP AI Launchpad 16. Prompt Registry 17. API Reference 18. Common Patterns 19. Troubleshooting 20. References

Overview

SAP AI Core is a service on SAP Business Technology Platform (BTP) that manages AI asset execution in a standardized, scalable, hyperscaler-agnostic manner. SAP AI Launchpad provides the management UI for AI runtimes including the Generative AI Hub.

Core Capabilities

Three Components

1. SAP AI Core: Execution engine for AI workflows and model serving 2. SAP AI Launchpad: Management UI for AI runtimes and GenAI Hub 3. AI API: Standardized lifecycle management across runtimes

Quick Start

Prerequisites

• SAP BTP enterprise account
• SAP AI Core service instance (Extended plan for GenAI)
• Service key with credentials

1. Get Authentication Token

# Set environment variables from service key
export AI_API_URL="<your-ai-api-url>"
export AUTH_URL="<your-auth-url>"
export CLIENT_ID="<your-client-id>"
export CLIENT_SECRET="<your-client-secret>"

# Get OAuth token
AUTH_TOKEN=$(curl -s -X POST "$AUTH_URL/oauth/token" \
  -H "Content-Type: application/x-www-form-urlencoded" \
  -d "grant_type=client_credentials&client_id=$CLIENT_ID&client_secret=$CLIENT_SECRET" \
  | jq -r '.access_token')

2. Create Orchestration Deployment

# Check for existing orchestration deployment
curl -X GET "$AI_API_URL/v2/lm/deployments" \
  -H "Authorization: Bearer $AUTH_TOKEN" \
  -H "AI-Resource-Group: default" \
  -H "Content-Type: application/json"

# Create orchestration deployment if needed
curl -X POST "$AI_API_URL/v2/lm/deployments" \
  -H "Authorization: Bearer $AUTH_TOKEN" \
  -H "AI-Resource-Group: default" \
  -H "Content-Type: application/json" \
  -d '{
    "configurationId": "<orchestration-config-id>"
  }'

3. Use Harmonized API for Model Inference

ORCHESTRATION_URL="<deployment-url>"

curl -X POST "$ORCHESTRATION_URL/v2/completion" \
  -H "Authorization: Bearer $AUTH_TOKEN" \
  -H "AI-Resource-Group: default" \
  -H "Content-Type: application/json" \
  -d '{
    "config": {
      "module_configurations": {
        "llm_module_config": {
          "model_name": "gpt-4o",
          "model_version": "latest",
          "model_params": {
            "max_tokens": 1000,
            "temperature": 0.7
          }
        },
        "templating_module_config": {
          "template": [
            {"role": "system", "content": "You are a helpful assistant."},
            {"role": "user", "content": "{{?user_query}}"}
          ]
        }
      }
    },
    "input_params": {
      "user_query": "What is SAP AI Core?"
    }
  }'

Service Plans

Key Restrictions:

• Free and Standard mutually exclusive in same subaccount
• Free → Standard upgrade possible; downgrade not supported
• Max 50 resource groups per tenant

Model Providers

SAP AI Core provides access to model providers through a tenant-specific catalog. Treat exact model names and versions as examples until verified in the target tenant with GET /v2/lm/scenarios/foundation-models/models or SAP AI Launchpad Model Library.

• Azure OpenAI: GPT-family chat, vision, reasoning, realtime, and embedding models where entitled
• SAP Open Source: Llama/Falcon/Mistral-family open source models where enabled
• Google Vertex AI: Gemini-family chat, vision, code, and embedding models where entitled
• AWS Bedrock: Anthropic Claude and Amazon model families where entitled
• Mistral AI: Mistral Large/Small/Codestral-family models where enabled
• IBM: Granite models
• Perplexity: Sonar-family web-grounded models where enabled

For detailed provider configurations and model lists, see references/model-providers.md.

Orchestration

The orchestration service provides unified access to multiple models through a modular pipeline with 8 execution stages: 1. Grounding → 2. Templating (mandatory) → 3. Input Translation → 4. Data Masking → 5. Input Filtering → 6. Model Configuration (mandatory) → 7. Output Filtering → 8. Output Translation

For complete orchestration module configurations, examples, and advanced patterns, see references/orchestration-modules.md.

Content Filtering

Azure Content Safety: Filters content across 4 categories (Hate, Violence, Sexual, SelfHarm) with severity levels 0-6. Azure OpenAI blocks severity 4+ automatically. Additional features include PromptShield and Protected Material detection.

Llama Guard 3: Covers 14 categories including violent crimes, privacy violations, and code interpreter abuse.

Data Masking

Two PII protection methods:

• Anonymization: MASKED_ENTITY (non-reversible)
• Pseudonymization: MASKED_ENTITY_ID (reversible)

Supported entities (25 total): Personal data, IDs, financial information, SAP-specific IDs, and sensitive attributes. For complete entity list and implementation details, see references/orchestration-modules.md.

Grounding (RAG)

Integrate external data from SharePoint, S3, SFTP, SAP Build Work Zone, and DMS. Supports PDF, HTML, DOCX, images, and more. Limit: 2,000 documents per pipeline with daily refresh. For detailed setup, see references/grounding-rag.md.

Tool Calling

Enable LLMs to execute functions through a 5-step workflow: define tools → receive tool_calls → execute functions → return results → LLM incorporates responses. Templates available in templates/tool-definition.json.

Structured Output

Force model responses to match JSON schemas using strict validation. Useful for structured data extraction and API responses.

Embeddings

Generate semantic embeddings for RAG and similarity search via /v2/embeddings endpoint. Supports document, query, and text input types.

ML Training

Uses Argo Workflows for training pipelines. Key requirements: create default object store secret, define workflow template, create configuration with parameters, and execute training. For complete workflow patterns, see references/ml-operations.md.

Deployments

Deploy models via two-step process: create configuration (with model binding), then create deployment with TTL. Statuses: Pending → Running → Stopping → Stopped/Dead. Templates in templates/deployment-config.json.

SAP AI Launchpad

Web-based UI with 4 key applications:

• Workspaces: Manage connections and resource groups
• ML Operations: Train, deploy, monitor models
• Generative AI Hub: Prompt experimentation and orchestration
• Functions Explorer: Explore available AI functions

Required roles include genai_manager, genai_experimenter, prompt_manager, orchestration_executor, and mloperations_editor. For complete guide, see references/ai-launchpad-guide.md.

Prompt Registry

The Prompt Registry manages the lifecycle of prompt templates from design to runtime, integrating them into SAP AI Core and orchestration workflows.

Two management interfaces:

• Imperative API: Full CRUD via REST, for design-time prompt refinement
• Declarative API: Git repository sync, for runtime and CI/CD use cases

Key endpoints:

• POST /v2/lm/promptTemplates — Create a prompt template
• POST /v2/lm/promptTemplates/{id}/substitution — Fill template by ID
• POST /v2/lm/scenarios/{scenario}/promptTemplates/{name}/versions/{version}/substitution — Fill by name

For complete Prompt Registry documentation, see references/ai-launchpad-guide.md.

API Reference

Core Endpoints

Key endpoints: /v2/lm/scenarios, /v2/lm/configurations, /v2/lm/deployments, /v2/lm/executions, /lm/meta. For complete API reference with examples, see references/api-reference.md.

Common Patterns

CAP Integration: SAP CAP is the primary consumer framework for AI Core on BTP. Bind an AI Core service instance to your CAP app via MTA, then call the orchestration API from CAP event handlers using the SAP Cloud SDK for AI. Always process LLM calls asynchronously in production (return 202 Accepted, process in background via cds.spawn) to avoid BTP load balancer timeouts. See sap-cap-capire and sap-cloud-sdk-ai skills for complete code examples.

Simple Chat: Basic model invocation with templating module RAG with Grounding: Combine vector search with LLM for context-aware responses Secure Enterprise Chat: Filtering + masking + grounding for PII protection Templates available in templates/orchestration-workflow.json.

Troubleshooting

Common Issues:

• 401 Unauthorized: Refresh OAuth token
• 403 Forbidden: Check IAM roles, request quota increase
• 404 Not Found: Verify AI-Resource-Group header
• Deployment DEAD: Check deployment logs
• Training failed: Create default object store secret

Request quota increases via support ticket (Component: CA-ML-AIC).

Bundled Resources

Reference Documentation

1. references/orchestration-modules.md - All orchestration modules in detail 2. references/generative-ai-hub.md - Complete GenAI hub documentation 3. references/model-providers.md - Model providers and configurations 4. references/api-reference.md - Complete API endpoint reference 5. references/grounding-rag.md - Grounding and RAG implementation 6. references/ml-operations.md - ML operations and training 7. references/advanced-features.md - Chat, applications, security, auditing 8. references/ai-launchpad-guide.md - Complete SAP AI Launchpad UI guide

Templates

1. templates/deployment-config.json - Deployment configuration template 2. templates/orchestration-workflow.json - Orchestration workflow template 3. templates/tool-definition.json - Tool calling definition template

Official Sources

• SAP AI Core Guide: https://help.sap.com/docs/sap-ai-core
• SAP AI Launchpad Guide: https://help.sap.com/docs/sap-ai-launchpad
• SAP Note 3437766: Model token rates and limits

SAP AI Core & AI Launchpad Skill

A portable AI coding assistant skill for SAP AI Core and SAP AI Launchpad development on SAP Business Technology Platform (BTP). Claude-specific command metadata is packaging support only; the skill content is intended to remain useful in Codex, OpenCode, and other Markdown-capable harnesses.

Capability Index

Overview

This skill provides guidance for:

• Deploying and consuming generative AI models
• Building orchestration workflows with templating, filtering, and grounding
• Implementing RAG (Retrieval-Augmented Generation) with vector databases
• Managing ML training pipelines with Argo Workflows
• Configuring content filtering and data masking for PII protection
• Using the Generative AI Hub for prompt experimentation

When to Use This Skill

This skill is triggered when working with:

SAP AI Core

• SAP AI Core setup and configuration
• SAP AI Core deployments and executions
• SAP AI Core service plans (Free, Standard, Extended)
• SAP AI Core API endpoints
• AI Core generative AI hub
• AI Core orchestration service
• AI Core foundation models
• AI Core grounding and RAG

SAP AI Launchpad

• SAP AI Launchpad setup
• AI Launchpad generative AI hub
• AI Launchpad prompt experimentation
• AI Launchpad ML operations
• AI Launchpad orchestration workflows

Model Providers

• Azure OpenAI on SAP
• GPT-4o, GPT-4 Turbo, GPT-3.5 on SAP
• AWS Bedrock on SAP
• Claude on SAP BTP
• Anthropic Claude via AI Core
• Google Vertex AI on SAP
• Gemini on SAP
• Mistral AI on SAP
• IBM Granite on SAP
• Llama models on SAP

Features

• LLM deployment on SAP
• Generative AI on SAP BTP
• AI model orchestration
• Prompt templating
• Content filtering for AI
• Data masking for AI
• PII protection in AI
• Vector database integration
• Document grounding
• RAG implementation SAP
• Embeddings generation
• Tool calling with LLMs
• Function calling AI Core
• Structured output AI
• JSON schema responses
• Streaming responses

ML Operations

• ML model training SAP
• Argo Workflows SAP
• Training pipelines
• Batch inference SAP
• Model deployment SAP
• Training schedules
• Execution management
• Artifact management

API & Integration

• AI API SAP
• Harmonized API
• Chat completion API
• Embeddings API
• Orchestration API
• REST API AI Core

Advanced Features

• Multi-turn chat conversations
• Git repository sync applications
• Prompt templates declarative
• Prompt optimization SAP
• AI content as a service
• AI content security
• Data protection privacy GDPR
• Auditing logging SAP AI
• KServe serving templates
• Metadata vector search
• Content packages DataRobot

Keywords

sap ai core, sap ai launchpad, generative ai hub, foundation models,
llm deployment, model orchestration, prompt templating, content filtering,
data masking, pii protection, vector database, document grounding, rag,
embeddings, tool calling, function calling, structured output, streaming,
azure openai sap, gpt-4 sap, claude sap, gemini sap, mistral sap,
llama sap, aws bedrock sap, google vertex ai sap, ibm granite sap,
ml operations, argo workflows, training pipeline, batch inference,
model deployment, ai api, harmonized api, orchestration api,
sap btp ai, enterprise ai, sap machine learning, ai core extended,
prompt experimentation, ai launchpad workspaces, resource groups,
configurations, executions, deployments, artifacts, scenarios,
chat conversations, messages history, git sync applications,
prompt templates, prompt optimization, ai content security,
data protection, gdpr compliance, auditing logging, kserve,
serving templates, metadata retrieval, content packages

File Structure

sap-ai-core/
├── SKILL.md                        # Main skill file
├── README.md                       # This file
├── references/
│   ├── orchestration-modules.md    # Detailed orchestration module docs
│   ├── generative-ai-hub.md        # Generative AI Hub reference
│   ├── api-reference.md            # Complete API reference
│   ├── grounding-rag.md            # Grounding and RAG implementation
│   ├── ml-operations.md            # ML training and operations
│   ├── model-providers.md          # Model providers and configurations
│   ├── advanced-features.md        # Chat, security, auditing, templates
│   └── ai-launchpad-guide.md       # Complete AI Launchpad UI guide
└── templates/
    ├── deployment-config.json      # Deployment configuration template
    ├── orchestration-workflow.json # Orchestration workflow template
    └── tool-definition.json        # Tool calling definition template

Prerequisites

• SAP BTP enterprise account
• SAP AI Core service instance
• Extended service plan (for Generative AI Hub)
• Service key with credentials

Quick Start

1. Set up authentication:

export AI_API_URL="<your-ai-api-url>"
export AUTH_TOKEN="<your-oauth-token>"

2. List available models:

curl -X GET "$AI_API_URL/v2/lm/scenarios/foundation-models/models" \
  -H "Authorization: Bearer $AUTH_TOKEN" \
  -H "AI-Resource-Group: default"

3. Create orchestration deployment and start using models

Before using any model ID from examples or references, list the target tenant catalog:

curl -X GET "$AI_API_URL/v2/lm/scenarios/foundation-models/models" \
  -H "Authorization: Bearer $AUTH_TOKEN" \
  -H "AI-Resource-Group: default"

Documentation Sources

License

GPL-3.0

Version

Current: 2.3.0 (2026-06-16)

• Documentation-audited AI Core guidance with tenant/runtime verification still pending.
• Model names are examples and must be checked against the target tenant catalog.
• Skill content is portable across Claude, Codex, OpenCode, and similar Markdown-capable harnesses.

Last Updated

2026-06-16

Next Review

Next source refresh and tenant verification remain pending until SAP Help/package evidence or live tenant evidence is available.

Advanced Features Reference

Complete reference for additional SAP AI Core features not covered in other reference files.

Documentation Source: https://github.com/SAP-docs/sap-artificial-intelligence/tree/main/docs/sap-ai-core

---

Table of Contents

1. Chat Conversations 2. Applications (Git Sync) 3. Prompt Templates 4. Prompt Optimization 5. AI Content as a Service 6. AI Content Security 7. Data Protection and Privacy 8. Auditing and Logging 9. ServingTemplate Schema 10. Contextualized Retrieval with Metadata 11. Content Packages

---

Chat Conversations

Multi-turn conversation handling using the orchestration service.

Message History Management

The orchestration service manages conversation history through the messages_history parameter, storing user and assistant role exchanges.

Request Structure

{
  "orchestration_config": {
    "module_configurations": {
      "templating_module_config": {
        "template": [
          {"role": "user", "content": "{{?current_message}}"}
        ]
      },
      "llm_module_config": {
        "model_name": "gpt-4o",
        "model_version": "latest",
        "model_params": {
          "max_tokens": 300,
          "temperature": 0.1
        }
      }
    }
  },
  "messages_history": [
    {"role": "user", "content": "What is machine learning?"},
    {"role": "assistant", "content": "Machine learning is a branch of AI..."},
    {"role": "user", "content": "Can you give an example?"},
    {"role": "assistant", "content": "Sure, email spam filtering is an example..."}
  ],
  "input_params": {
    "current_message": "What about deep learning?"
  }
}

Key Behavior

• The templating module appends the current user message to the message history
• The combined history generates the prompt sent to the LLM module
• Response module_results.templating and orchestration_result.choices can be used as message history for subsequent requests

Continuation Pattern

def continue_conversation(history, new_message, response):
    """Update conversation history with new exchange."""
    history.append({"role": "user", "content": new_message})
    history.append({
        "role": "assistant",
        "content": response["orchestration_result"]["choices"][0]["message"]["content"]
    })
    return history

---

Applications (Git Sync)

Applications synchronize workflow templates from GitHub repositories.

Key Features

• Automatic Sync: Applications sync with GitHub every ~3 minutes
• Manual Sync: Trigger via POST {{apiurl}}/admin/applications/{{appName}}/refresh

API Endpoints

Create Application

curl -X POST "$AI_API_URL/v2/admin/applications" \
  -H "Authorization: Bearer $AUTH_TOKEN" \
  -H "AI-Resource-Group: default" \
  -H "Content-Type: application/json" \
  -d '{
    "applicationName": "my-workflows",
    "repositoryUrl": "https://github.com/org/ai-workflows",
    "revision": "HEAD",
    "path": "workflows/"
  }'

Required Configuration

Sync Status Response

{
  "health": "healthy",
  "lastSyncTime": "2024-01-15T10:00:00Z",
  "status": "Synced",
  "message": ""
}

Validation Checks

The system validates:

• No duplicate workflow names
• Correct scenario labels on templates
• Valid YAML syntax
• Proper metadata structure (WorkflowTemplate kind)

---

Prompt Templates

Manage prompts through declarative (Git) or imperative (API) approaches.

Declarative Approach (Git-managed)

File Format

Filename: <name>.prompttemplate.ai.sap.yaml

name: customer-support-prompt
version: 0.0.1
scenario: customer-service
spec:
  template:
    - role: system
      content: "You are a helpful customer support agent for {{?company_name}}."
    - role: user
      content: "{{?customer_query}}"
defaults:
  company_name: "Acme Corp"
additional_fields:
  metadata:
    author: "AI Team"
    category: "support"
  model_restrictions:
    blocked_models:
      - model_name: "gpt-3.5-turbo"
        versions: ["0613"]

Key Characteristics

• Managed through git commits
• Auto-sync with prompt registry
• Marked as managedBy: declarative
• Always reflects HEAD version
• Cannot be edited via imperative API

Imperative Approach (API-managed)

Create Prompt Template

curl -X POST "$AI_API_URL/v2/lm/promptTemplates" \
  -H "Authorization: Bearer $AUTH_TOKEN" \
  -H "AI-Resource-Group: default" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "api-managed-prompt",
    "version": "1.0.0",
    "scenario": "foundation-models",
    "spec": {
      "template": [
        {"role": "user", "content": "{{?user_input}}"}
      ]
    }
  }'

List Prompt Templates

curl -X GET "$AI_API_URL/v2/lm/promptTemplates" \
  -H "Authorization: Bearer $AUTH_TOKEN" \
  -H "AI-Resource-Group: default"

Placeholder Syntax

---

Prompt Optimization

Automated prompt improvement using optimization runs.

Overview

Prompt Optimization takes an input prompt template and a dataset of desirable responses to maximize a specified metric.

Prerequisites

• Required roles: genai_manager or custom_evaluation
• Service plan: extended tier required
• Object store named default must be registered
• Prompt template saved in prompt registry
• Dataset artifact prepared and registered

Dataset Preparation

Dataset Structure

[
  {
    "fields": {
      "input": "Customer complaint about delivery delay",
      "category": "logistics"
    },
    "answer": {
      "sentiment": "negative",
      "urgency": "high",
      "category": "delivery"
    }
  }
]

Important: Placeholder names in fields must match those in the template exactly. Do not include confidential or personally identifiable information.

Process

1. Submit optimization job with prompt template and dataset 2. System generates multiple prompt variations 3. Evaluates variations against target metric 4. Returns optimized prompt to registry 5. Stores additional results in object store

Launchpad UI Flow (7 Steps)

1. Access: Connect to SAP AI Core via Workspaces app 2. Navigate: Generative AI Hub → Optimization 3. Initiate: Create → Prompt Optimization 4. Configure Artifacts: Select template, models, and dataset 5. Select Metric: Choose evaluation metric 6. Advanced Settings: Configure template name/version (optional) 7. Review & Deploy: Verify and start job

Limitations

Operations (AI Launchpad)

• Create a new prompt optimization
• View existing prompt optimizations
• View detailed run information

---

AI Content as a Service

Publish AI content to SAP BTP Service Marketplace.

Capabilities

• Publish workflows, serving templates, or Docker images
• Distribute as managed service on SAP BTP
• Other tenants can consume via standard APIs

Use Cases

• Monetize AI models and workflows
• Share AI components across organization
• Provide standardized AI services

---

AI Content Security

Security best practices for AI content (workflows, templates, Docker images).

Required Practices

Key Responsibility

"Users of AI Core are responsible for the content of their Docker images and assume the risk of running compromised containers in the platform."

Docker Security Guidelines

1. Select minimal, hardened base images 2. Keep images updated 3. Remove unnecessary components 4. Use multi-stage builds 5. Scan images for vulnerabilities 6. Sign images for verification

---

Data Protection and Privacy

Compliance features for data protection.

Supported Capabilities

Compliance Scope

• General data protection acts (GDPR, etc.)
• Industry-specific legislation
• Regional privacy requirements

Important Notes

• SAP does not provide legal advice
• Compliance requires secure system operation
• Case-by-case evaluation required

---

Auditing and Logging

Security event logging in SAP AI Core.

Events Logged

Log Details by Operation Type

Authentication Events

---

ServingTemplate Schema

API schema for serving templates (KServe integration) for model deployment.

Quotas and Limits

Resource Structure

apiVersion: ai.sap.com/v1alpha1
kind: ServingTemplate
metadata:
  name: my-serving-template
  annotations:
    scenarios.ai.sap.com/description: "Description of scenario"
    scenarios.ai.sap.com/name: "scenario-name"
    executables.ai.sap.com/description: "Description of executable"
    executables.ai.sap.com/name: "executable-name"
    ai.sap.com/bulkUpdates: "true"  # Enable bulk operations
  labels:
    ai.sap.com/version: "1.0.0"
    scenarios.ai.sap.com/id: "unique-scenario-id"
spec:
  inputs:
    parameters:
      - name: modelUri
        default: ""
        type: string
    artifacts:
      - name: model
  template:
    apiVersion: serving.kserve.io/v1beta1
    metadata:
      name: "{{inputs.parameters.name}}"
    spec:
      predictor:
        containers:
          - name: kserve-container
            image: "{{inputs.parameters.image}}"
            env:
              - name: STORAGE_URI
                value: "{{inputs.artifacts.model}}"

Model Path Configuration

Important: The /mnt/models path is the typical default used in SAP AI Core examples, but the mount path is configurable via the ServingRuntime/ServingTemplate and container args (e.g., --model_dir). Your inference code should read the path from configuration or environment variables rather than assuming a hardcoded path:

import os

# Read from environment or use default
MODEL_PATH = os.environ.get("MODEL_DIR", "/mnt/models")

def load_model():
    """Load model from configured mount path."""
    return load_from_path(MODEL_PATH)

Configuration Options:

• Set via container args: --model_dir=/custom/path
• Set via environment variable in ServingTemplate
• Override in KServe InferenceService spec

Annotations Reference

Labels Reference

Parameter Types

Only string type is supported for input parameters.

Placeholder Syntax

Use {{inputs.parameters.ParameterName}} and {{inputs.artifacts.ArtifactName}} in template spec.

Bulk Operations

When ai.sap.com/bulkUpdates: "true" is set:

# Bulk stop deployments
curl -X PATCH "$AI_API_URL/v2/lm/deployments" \
  -H "Authorization: Bearer $AUTH_TOKEN" \
  -H "AI-Resource-Group: default" \
  -H "Content-Type: application/json" \
  -d '{
    "targetStatus": "STOPPED",
    "deploymentIds": ["deploy-1", "deploy-2", "deploy-3"]
  }'

# Bulk delete deployments
curl -X DELETE "$AI_API_URL/v2/lm/deployments" \
  -H "Authorization: Bearer $AUTH_TOKEN" \
  -H "AI-Resource-Group: default" \
  -H "Content-Type: application/json" \
  -d '{
    "deploymentIds": ["deploy-1", "deploy-2", "deploy-3"]
  }'

---

Contextualized Retrieval with Metadata

Include metadata in grounding retrieval results.

Configuration

Add metadata_params to grounding configuration:

{
  "grounding_module_config": {
    "grounding_service": "document_grounding_service",
    "grounding_service_configuration": {
      "grounding_input_parameters": ["user_query"],
      "grounding_output_parameter": "context",
      "metadata_params": ["source", "webUrl", "title"],
      "filters": [{"id": "<pipeline-id>"}]
    }
  }
}

Metadata Levels

Discovery

Query available metadata keys:

curl -X POST "$AI_API_URL/v2/lm/document-grounding/retrieval/search" \
  -H "Authorization: Bearer $AUTH_TOKEN" \
  -H "AI-Resource-Group: default" \
  -H "Content-Type: application/json" \
  -d '{
    "query": "test query",
    "filters": [{"id": "<pipeline-id>"}]
  }'

Naming Convention for Conflicts

When metadata keys exist at multiple levels:

• Chunk-level: webUrl
• Document-level: document_webUrl
• Repository-level: repository_webUrl

Using Metadata in Prompts

{
  "templating_module_config": {
    "template": [
      {
        "role": "system",
        "content": "Answer based on the context. Include source references.\n\nContext: {{$context}}"
      },
      {"role": "user", "content": "{{?user_query}}"}
    ]
  }
}

---

Content Packages

Additional Python packages extending SAP AI Core.

Available Packages

Installation

pip install sap-ai-core-datarobot
pip install sap-computer-vision-package

Computer Vision Package Capabilities

• Image classification
• Feature extraction
• Integration with SAP AI SDK Core

---

Documentation Links

• Chat: https://github.com/SAP-docs/sap-artificial-intelligence/blob/main/docs/sap-ai-core/chat-39321a9.md
• Applications: https://github.com/SAP-docs/sap-artificial-intelligence/blob/main/docs/sap-ai-core/application-7f1e35b.md
• Prompt Templates: https://github.com/SAP-docs/sap-artificial-intelligence/blob/main/docs/sap-ai-core/create-a-prompt-template-declarative-815def5.md
• AI Content Security: https://github.com/SAP-docs/sap-artificial-intelligence/blob/main/docs/sap-ai-core/ai-content-security-d1cd77f.md
• Data Protection: https://github.com/SAP-docs/sap-artificial-intelligence/blob/main/docs/sap-ai-core/data-protection-and-privacy-d25e4c9.md
• Auditing: https://github.com/SAP-docs/sap-artificial-intelligence/blob/main/docs/sap-ai-core/auditing-and-logging-information-e19844a.md
• API Schema: https://github.com/SAP-docs/sap-artificial-intelligence/blob/main/docs/sap-ai-core/api-schema-spec-ai-sap-com-v1alpha1-4d1ffd2.md

SAP AI Launchpad Complete Guide

Comprehensive reference for SAP AI Launchpad features and operations.

Documentation Source: https://github.com/SAP-docs/sap-artificial-intelligence/tree/main/docs/sap-ai-launchpad

---

Table of Contents

1. Overview 2. Initial Setup 3. Workspaces and Connections 4. User Roles 5. Generative AI Hub 6. Prompt Editor 7. Prompt Registry 8. Prompt Optimization 9. Orchestration Workflows 10. ML Operations 11. Configurations 12. Deployments 13. Executions and Runs 14. Schedules 15. Datasets and Artifacts 16. Model Comparison 17. Applications 18. Meta API and Custom Runtime Capabilities

---

Overview

SAP AI Launchpad is a multitenant SaaS application on SAP BTP that provides:

• Management UI for AI runtimes (SAP AI Core)
• Generative AI Hub for prompt experimentation
• ML Operations for model lifecycle management
• Analytics and monitoring dashboards

Two User Types

---

Initial Setup

Prerequisites

1. SAP BTP enterprise account 2. Subaccount with Cloud Foundry enabled 3. SAP AI Launchpad subscription 4. SAP AI Core instance (for runtime connection)

Setup Steps

1. Create Subaccount with Cloud Foundry environment 2. Subscribe to SAP AI Launchpad in Service Marketplace 3. Create Service Instance of SAP AI Core (if needed) 4. Assign Role Collections to users 5. Add Connection to SAP AI Core runtime

Service Plans

Note: Free → Standard upgrade preserves data; downgrade not supported.

---

Workspaces and Connections

Adding a Connection

1. Navigate to Administration → Connections 2. Click Add 3. Enter connection details:

• Name
• Service Key (from SAP AI Core)

4. Test connection 5. Save

Managing Connections

Assigning Connection to Workspace

1. Navigate to Workspaces 2. Select workspace 3. Click Assign Connection 4. Select connection from dropdown 5. Confirm

---

User Roles

Administrative Roles

ML Operations Roles

Generative AI Hub Roles

Functions Explorer Roles

Note: Role names prompt_media_executor and orchestration_executor may be deprecated. Verify current role names in SAP documentation.

---

Generative AI Hub

Access Path

Workspaces → Select workspace → Generative AI Hub

Features

Model Library

View model specifications including:

• Capabilities (chat, embeddings, vision)
• Context window sizes
• Performance benchmarks
• Cost per token
• Deprecation dates

---

Prompt Editor

Access

Generative AI Hub → Prompt Editor

Interface Elements

Message Roles

• System: Instructions for the model
• User: User input
• Assistant: Previous assistant responses

Variable Syntax

Use {{variable_name}} for placeholders with definitions section.

Running Prompts

1. Configure messages and variables 2. Select model (optional - uses default) 3. Adjust parameters 4. Click Run 5. View response (streaming available)

Image Inputs

• Supported for select models (GPT-4o, Gemini, Llama Vision)
• Maximum 5MB across all inputs
• Requires prompt_media_executor role

Saving Prompts

• Click Save (manager roles only)
• Assign to collection
• Add tags and notes
• Version automatically managed

Prompt Types

---

Prompt Registry

Access: Generative AI Hub → Prompt Registry

The Prompt Registry manages the lifecycle of prompt templates from design to runtime. It integrates prompt templates into SAP AI Core, making them discoverable across applications and orchestration workflows.

Two Management Interfaces

Imperative API: Create a Prompt Template

Send a POST request to {{apiurl}}/v2/lm/promptTemplates:

{
  "name": "summarization-template",
  "scenarioId": "foundation-models",
  "version": "0.0.1",
  "template": [
    {"role": "system", "content": "Summarize the following text concisely."},
    {"role": "user", "content": "{{?input_text}}"}
  ],
  "defaults": {
    "input_text": ""
  },
  "additional_fields": {}
}

Declarative API: Create via Git

1. Create a prompt template file: <name>.prompttemplate.ai.sap.yaml 2. Push to a synced git repository 3. The template auto-syncs within minutes 4. Verify via GET {{apiurl}}/v2/lm/promptTemplates 5. Declarative templates are marked managedBy: <declarative> and are always the head version

Using a Prompt Template at Runtime

Fill a template by ID:

POST {{apiurl}}/v2/lm/promptTemplates/{{promptTemplateId}}/substitution

Or fill by name, scenario, and version:

POST {{apiurl}}/v2/lm/scenarios/{{scenarioId}}/promptTemplates/{{promptTemplateName}}/versions/{{versionId}}/substitution

Include variable values in the request body. Add query parameter metadata=true to return the full template definition.

Resource Group Scope

By default, prompt templates are managed at the main-tenant level. To scope to a resource group:

--header 'AI-Resource-Group-Scope: true'
--header 'AI-Resource-Group: <resource group>'

Integration with Orchestration

Prompt templates from the registry can be referenced in orchestration workflows via the templating module, enabling centralized prompt management across multiple applications.

Required Roles

---

Prompt Optimization

Access: Generative AI Hub → Prompt Optimization

Prompt Optimization evaluates and refines prompts using reference datasets. Available since Q1 2026.

Key Features

• Variable mapping to reconcile mismatches between prompts and datasets
• Metrics including EXACT_MATCH (Boolean: output exactly matches reference)
• Automated evaluation against ground-truth responses

Variable Mapping

When prompt variable names differ from dataset column names, variable mapping aligns them:

{
  "variable_mapping": [
    {"prompt_variable": "user_query", "dataset_column": "question"},
    {"prompt_variable": "context_text", "dataset_column": "document"}
  ]
}

Configuration

Create a configuration for prompt optimization via the API, referencing a prompt template from the Prompt Registry and a dataset in the object store.

---

Orchestration Workflows

Access

Generative AI Hub → Orchestration → Create

Workflow Modules

Required Modules Explained:

• Templating: Constructs the actual prompt/messages sent to the LLM using input variables and context
• Model Configuration: Specifies which LLM model to use and its parameters (temperature, max_tokens, etc.)

Building Workflows

1. Click Create to start new workflow 2. Configure required modules (Templating, Model) 3. Enable optional modules via Edit 4. Configure each enabled module 5. Click Test to run workflow 6. Click Save to store configuration

JSON Upload

• Maximum file size: 200 KB
• Format: JSON with module_configurations
• Note: Workflows with images can be downloaded but not uploaded

Saving Workflows

• Save as configuration for reuse
• Assign name and description
• Link to deployments

---

ML Operations

Access

Workspaces → Select workspace → ML Operations

Components

---

Configurations

Creating Configuration

1. Navigate to ML Operations → Configurations 2. Click Create 3. Enter details:

• Name
• Scenario
• Executable
• Parameters
• Input artifacts

4. Save

Configuration Contents

---

Deployments

Creating Deployment

1. Navigate to ML Operations → Deployments 2. Click Create 3. Select configuration 4. Set duration (optional TTL) 5. Click Create

Deployment Details

Deployment Statuses

Operations

Bulk Operations

• Stop multiple deployments
• Delete multiple deployments (up to 100)

---

Executions and Runs

Creating Execution

1. Navigate to ML Operations → Executions 2. Click Create 3. Select configuration 4. Click Create

Execution Statuses

Viewing Execution Details

• Parameters and artifacts
• Status and timing
• Logs from pipeline
• Output artifacts
• Metrics

Comparing Executions

1. Select multiple executions 2. Click Compare 3. View side-by-side:

• Parameters
• Metrics
• Durations

4. Create charts for visualization

---

Schedules

Creating Schedule

1. Navigate to ML Operations → Schedules 2. Click Create 3. Select configuration 4. Set cron expression 5. Define start/end dates 6. Save

Cron Expression Format

┌───────── minute (0-59)
│ ┌─────── hour (0-23)
│ │ ┌───── day of month (1-31)
│ │ │ ┌─── month (1-12)
│ │ │ │ ┌─ day of week (0-6)
│ │ │ │ │
* * * * *

Schedule Operations

---

Datasets and Artifacts

Dataset Registration

1. Navigate to ML Operations → Datasets 2. Click Register 3. Enter details:

• Name
• URL (ai://secret-name/path)
• Scenario
• Description

4. Save

Artifact Types

Finding Artifacts

• Filter by scenario
• Search by name
• Sort by date
• View details

---

Model Comparison

Comparing Models

1. Navigate to ML Operations → Models 2. Select multiple models 3. Click Compare 4. View:

• Configuration differences
• Metric comparisons
• Performance charts

Creating Comparison Charts

1. Select metrics to compare 2. Choose chart type 3. Configure axes 4. Generate visualization

---

Applications

Managing Applications

Access: Administration → Applications

Operations

Chat Application

Create chat interfaces using deployed models:

1. Create application 2. Configure model deployment 3. Set disclaimer (optional) 4. Share application URL

---

Meta API and Custom Runtime Capabilities

The Meta API identifies which capabilities apply to a given AI runtime, allowing SAP AI Launchpad to display only relevant features.

Purpose

Supported Capabilities

Metadata Refresh

• Automatic: Refreshed periodically on schedule
• On-demand: Users can trigger manual refresh
• Administration: SAP Runtime team manages active capabilities

Custom Runtime Usage

Custom runtimes can selectively implement only necessary capabilities, creating a tailored experience:

AI Runtime → Meta API Query → Capability List → Filtered UI

---

Accessibility Features

SAP AI Launchpad provides:

• Keyboard navigation
• Screen reader support
• High contrast themes
• Accessible UI components

---

Language Settings

Change interface language: 1. Navigate to user settings 2. Select language preference 3. Save changes

Supported languages vary by region and deployment.

---

Documentation Links

• What is AI Launchpad: https://github.com/SAP-docs/sap-artificial-intelligence/blob/main/docs/sap-ai-launchpad/what-is-sap-ai-launchpad-760889a.md
• Initial Setup: https://github.com/SAP-docs/sap-artificial-intelligence/blob/main/docs/sap-ai-launchpad/initial-setup-5d8adb6.md
• Service Plans: https://github.com/SAP-docs/sap-artificial-intelligence/blob/main/docs/sap-ai-launchpad/service-plans-ec1717d.md
• ML Operations: https://github.com/SAP-docs/sap-artificial-intelligence/blob/main/docs/sap-ai-launchpad/ml-operations-df78271.md
• Generative AI Hub: https://github.com/SAP-docs/sap-artificial-intelligence/blob/main/docs/sap-ai-launchpad/generative-ai-hub-b0b935b.md
• Prompt Experimentation: https://github.com/SAP-docs/sap-artificial-intelligence/blob/main/docs/sap-ai-launchpad/prompt-experimentation-384cc0c.md
• Orchestration: https://github.com/SAP-docs/sap-artificial-intelligence/blob/main/docs/sap-ai-launchpad/build-your-orchestration-workflow-b7dc8b4.md
• Deployments: https://github.com/SAP-docs/sap-artificial-intelligence/blob/main/docs/sap-ai-launchpad/deployments-0543c2c.md

SAP AI Core API Reference

Complete API reference for SAP AI Core.

Documentation Source: https://github.com/SAP-docs/sap-artificial-intelligence/tree/main/docs/sap-ai-core

---

Authentication

OAuth Token Endpoint

curl -X POST "https://<id-zone>.authentication.<region>.hana.ondemand.com/oauth/token" \
  -H "Content-Type: application/x-www-form-urlencoded" \
  -d "grant_type=client_credentials&client_id=$CLIENT_ID&client_secret=$CLIENT_SECRET"

Required Headers

---

Base URLs

Regions: eu10, eu11, us10, us21, jp10, ap10, ap11

---

API Versioning

All endpoints use the /v2/* versioned routes:

• /v2/lm/* - Language model operations
• /v2/inference/* - Inference deployments
• /v2/admin/* - Administrative operations (secrets, repositories)

---

Scenarios

List Scenarios

GET $AI_API_URL/v2/lm/scenarios

Response:

{
  "count": 2,
  "resources": [
    {
      "id": "foundation-models",
      "name": "Foundation Models",
      "description": "Access to generative AI models"
    },
    {
      "id": "orchestration",
      "name": "Orchestration",
      "description": "Unified model access with pipeline features"
    }
  ]
}

---

Models

List Available Models

GET $AI_API_URL/v2/lm/scenarios/foundation-models/models

Response:

{
  "count": 50,
  "resources": [
    {
      "model": "gpt-4o",
      "accessType": "Remote",
      "displayName": "GPT-4o",
      "provider": "azure-openai",
      "executableId": "azure-openai",
      "versions": [
        {
          "name": "2024-05-13",
          "isLatest": true,
          "capabilities": ["text-generation", "chat"],
          "contextLength": 128000,
          "inputCost": 5.0,
          "outputCost": 15.0,
          "isStreamingSupported": true
        }
      ]
    }
  ]
}

---

Configurations

Create Configuration

POST $AI_API_URL/v2/lm/configurations

Request Body:

{
  "name": "my-config",
  "executableId": "azure-openai",
  "scenarioId": "foundation-models",
  "parameterBindings": [
    {"key": "modelName", "value": "gpt-4o"},
    {"key": "modelVersion", "value": "latest"}
  ],
  "inputArtifactBindings": []
}

Response:

{
  "id": "abc123-def456-ghi789",
  "message": "Configuration created"
}

List Configurations

GET $AI_API_URL/v2/lm/configurations

Query Parameters:

Get Configuration

GET $AI_API_URL/v2/lm/configurations/{configurationId}

Delete Configuration

DELETE $AI_API_URL/v2/lm/configurations/{configurationId}

---

Deployments

Create Deployment

POST $AI_API_URL/v2/lm/deployments

Request Body:

{
  "configurationId": "<configuration-id>",
  "ttl": "24h"
}

TTL Format: Natural numbers with units: m (minutes), h (hours), d (days)

• Valid: 5m, 2h, 7d
• Invalid: 4.5h, 4h30m (fractional and combined units not supported)
• Tip: Convert combined durations to a single unit (e.g., 270m instead of 4h30m)

Response:

{
  "id": "d12345-abcd-efgh",
  "deploymentUrl": "https://...",
  "status": "PENDING",
  "message": "Deployment created"
}

Get Deployment

GET $AI_API_URL/v2/lm/deployments/{deploymentId}

Response:

{
  "id": "d12345-abcd-efgh",
  "configurationId": "c12345-abcd",
  "configurationName": "my-config",
  "scenarioId": "foundation-models",
  "status": "RUNNING",
  "statusMessage": "",
  "deploymentUrl": "https://...",
  "createdAt": "2024-01-15T10:00:00Z",
  "modifiedAt": "2024-01-15T10:05:00Z"
}

Deployment Statuses

List Deployments

GET $AI_API_URL/v2/lm/deployments

Update Deployment

PATCH $AI_API_URL/v2/lm/deployments/{deploymentId}

Request Body:

{
  "configurationId": "<new-configuration-id>"
}

Stop Deployment

PATCH $AI_API_URL/v2/lm/deployments/{deploymentId}

Request Body:

{
  "targetStatus": "STOPPED"
}

Delete Deployment

DELETE $AI_API_URL/v2/lm/deployments/{deploymentId}

---

Executions

Create Execution

POST $AI_API_URL/v2/lm/executions

Request Body:

{
  "configurationId": "<configuration-id>"
}

Get Execution

GET $AI_API_URL/v2/lm/executions/{executionId}

Execution Statuses

List Executions

GET $AI_API_URL/v2/lm/executions

Stop Execution

PATCH $AI_API_URL/v2/lm/executions/{executionId}

Request Body:

{
  "targetStatus": "STOPPED"
}

Delete Execution

DELETE $AI_API_URL/v2/lm/executions/{executionId}

Get Execution Logs

GET $AI_API_URL/v2/lm/executions/{executionId}/logs

---

Artifacts

Register Artifact

POST $AI_API_URL/v2/lm/artifacts

Request Body:

{
  "name": "training-data",
  "kind": "dataset",
  "url": "ai://<object-store>/<path>",
  "scenarioId": "<scenario-id>",
  "description": "Training dataset"
}

Artifact Kinds:

• dataset: Training data
• model: Trained model
• resultset: Inference results
• other: Other artifacts

Get Artifact

GET $AI_API_URL/v2/lm/artifacts/{artifactId}

List Artifacts

GET $AI_API_URL/v2/lm/artifacts

---

Resource Groups

Create Resource Group

POST $AI_API_URL/v2/admin/resourceGroups

Request Body:

{
  "resourceGroupId": "my-resource-group"
}

List Resource Groups

GET $AI_API_URL/v2/admin/resourceGroups

Delete Resource Group

DELETE $AI_API_URL/v2/admin/resourceGroups/{resourceGroupId}

---

Secrets

Create Generic Secret

POST $AI_API_URL/v2/admin/secrets

Request Body:

{
  "name": "my-secret",
  "data": {
    "key1": "value1",
    "key2": "value2"
  }
}

Create Object Store Secret

POST $AI_API_URL/v2/admin/objectStoreSecrets

AWS S3:

{
  "name": "default",
  "type": "S3",
  "pathPrefix": "my-bucket/path",
  "data": {
    "AWS_ACCESS_KEY_ID": "<key>",
    "AWS_SECRET_ACCESS_KEY": "<secret>"
  }
}

List Secrets

GET $AI_API_URL/v2/admin/secrets

Delete Secret

DELETE $AI_API_URL/v2/admin/secrets/{secretName}

---

Meta API

Get Runtime Capabilities

GET $AI_API_URL/lm/meta

Response:

{
  "capabilities": {
    "logs.executions": true,
    "logs.deployments": true,
    "multitenant": true,
    "shareable": false,
    "staticDeployments": true,
    "userDeployments": true,
    "userExecutions": true,
    "timeToLiveDeployments": true,
    "analytics": true,
    "bulkUpdates": true,
    "executionSchedules": true
  },
  "limits": {
    "deployments.maxRunningCount": 10,
    "executions.maxRunningCount": 10,
    "minimumFrequencyHour": 1,
    "timeToLiveDeployments.minimum": "5m",
    "timeToLiveDeployments.maximum": "90d"
  },
  "extensions": {
    "analytics": "1.0",
    "metrics": "1.0",
    "resourceGroups": "1.0",
    "dataset": "1.0"
  }
}

---

Orchestration API

Chat Completion

POST $ORCHESTRATION_URL/v2/completion

Request Body:

{
  "config": {
    "module_configurations": {
      "llm_module_config": {
        "model_name": "gpt-4o",
        "model_version": "latest",
        "model_params": {
          "max_tokens": 1000,
          "temperature": 0.7
        }
      },
      "templating_module_config": {
        "template": [
          {"role": "system", "content": "{{?system}}"},
          {"role": "user", "content": "{{?user}}"}
        ]
      }
    }
  },
  "input_params": {
    "system": "You are a helpful assistant.",
    "user": "Hello!"
  }
}

Streaming Completion

POST $ORCHESTRATION_URL/v2/completion

Request Body:

{
  "config": {
    "module_configurations": {
      "llm_module_config": {
        "model_name": "gpt-4o",
        "model_version": "latest",
        "model_params": {
          "stream": true
        }
      },
      "templating_module_config": {
        "template": [{"role": "user", "content": "{{?prompt}}"}]
      }
    }
  },
  "input_params": {"prompt": "Tell me a story"}
}

Embeddings

POST $ORCHESTRATION_URL/v2/embeddings

Request Body:

{
  "config": {
    "module_configurations": {
      "embedding_module_config": {
        "model_name": "text-embedding-3-large",
        "model_version": "latest",
        "model_params": {
          "encoding_format": "float",
          "dimensions": 1024
        }
      }
    }
  },
  "input": ["Text to embed"]
}

---

Grounding API

Create Pipeline

POST $AI_API_URL/v2/lm/groundingPipelines

Request Body (SharePoint):

{
  "name": "hr-docs-pipeline",
  "configuration": {
    "dataSource": {
      "type": "sharepoint",
      "configuration": {
        "siteUrl": "https://company.sharepoint.com/sites/HR",
        "folderPath": "/Documents/Policies"
      }
    },
    "secretName": "sharepoint-secret"
  }
}

List Pipelines

GET $AI_API_URL/v2/lm/groundingPipelines

Delete Pipeline

DELETE $AI_API_URL/v2/lm/groundingPipelines/{pipelineId}

---

Error Responses

Standard Error Format

{
  "error": {
    "code": "ERROR_CODE",
    "message": "Human-readable error message",
    "requestId": "req-12345",
    "target": "deployments"
  }
}

Common Error Codes

---

Bulk Operations

Bulk Update Deployments

PATCH $AI_API_URL/v2/lm/deployments

Request Body:

{
  "deployments": [
    {"id": "dep1", "targetStatus": "STOPPED"},
    {"id": "dep2", "targetStatus": "STOPPED"}
  ]
}

Limit: 100 items per request

Bulk Delete Deployments

DELETE $AI_API_URL/v2/lm/deployments

Request Body:

{
  "deploymentIds": ["dep1", "dep2", "dep3"]
}

---

Schedules

Create Schedule

POST $AI_API_URL/v2/lm/executionSchedules

Request Body:

{
  "configurationId": "<config-id>",
  "cron": "0 0 * * *",
  "start": "2024-01-01T00:00:00Z",
  "end": "2024-12-31T23:59:59Z"
}

Cron Format: minute hour day month weekday

List Schedules

GET $AI_API_URL/v2/lm/executionSchedules

Delete Schedule

DELETE $AI_API_URL/v2/lm/executionSchedules/{scheduleId}

---

Documentation Links

• AI API Overview: https://github.com/SAP-docs/sap-artificial-intelligence/blob/main/docs/sap-ai-core/ai-api-overview-716d4c3.md
• Configurations: https://github.com/SAP-docs/sap-artificial-intelligence/blob/main/docs/sap-ai-core/create-configurations-884ae34.md
• Deployments: https://github.com/SAP-docs/sap-artificial-intelligence/blob/main/docs/sap-ai-core/deploy-models-dd16e8e.md

Generative AI Hub Reference

Complete reference for SAP AI Core Generative AI Hub.

Documentation Source: SAP Help Portal - SAP AI Core

---

Overview

The Generative AI Hub integrates large language models (LLMs) into SAP AI Core and SAP AI Launchpad, providing unified access to models from multiple providers.

Model availability note: SAP AI Core model IDs, versions, regions, and deprecation dates change by tenant, service plan, entitlement, and SAP Note 3437766 updates. Treat model names in this reference as examples. Before creating deployments, verify the exact target catalog in SAP AI Launchpad Model Library or with GET /v2/lm/scenarios/foundation-models/models.

Key Features

• Access to LLMs from multiple providers via unified API
• Harmonized API for model switching without code changes
• Prompt experimentation in AI Launchpad UI
• Prompt Registry for prompt template lifecycle management (available since Q1 2026)
• Orchestration workflows with filtering, masking, grounding, translation
• Token-based metering and billing

Prerequisites

• SAP AI Core with Extended service plan
• Valid service key credentials
• Resource group created

---

Global Scenarios

Two scenarios provide generative AI access:

---

Model Providers

1. Azure OpenAI (azure-openai)

Access to OpenAI models via Azure's private instance.

Example model families to verify in tenant catalog:

• GPT-family chat and multimodal models
• Reasoning model families
• Realtime conversational models, where enabled
• Text embedding models

Deprecated/retiring patterns: older GPT-4, GPT-4 Turbo, GPT-4-32k, and GPT-3.5-era deployments should be checked against SAP Note 3437766 and migrated before retirement dates shown in the tenant catalog.

Capabilities: Chat, embeddings, vision, reasoning, realtime

2. SAP-Hosted Open Source (aicore-opensource)

SAP-hosted open source models via OpenAI-compatible API.

Example model families to verify in tenant catalog:

• Llama-family chat and vision models
• Mistral/Mixtral-family instruction models
• Falcon-family models

Capabilities: Chat, embeddings, vision (select models)

3. Google Vertex AI (gcp-vertexai)

Access to Google's AI models.

Example model families to verify in tenant catalog:

• Gemini-family chat, vision, code, and long-context models
• Gemini Flash-family lower-latency models
• Google embedding models

Deprecated/retiring patterns: older Gemini and PaLM-era deployments should be checked against SAP Note 3437766 and migrated before retirement dates shown in the tenant catalog.

Capabilities: Chat, embeddings, vision, code, image generation

4. AWS Bedrock (aws-bedrock)

Access to models via AWS Bedrock.

Example model families to verify in tenant catalog:

• Anthropic Claude-family chat models
• Amazon Nova-family models
• Amazon Titan text and embedding models

Capabilities: Chat, embeddings

5. Mistral AI (aicore-mistralai)

SAP-hosted Mistral models.

Models:

• Mistral Large
• Mistral Medium
• Mistral Small
• Mistral 7B Instruct
• Codestral

Capabilities: Chat, code

6. IBM (aicore-ibm)

SAP-hosted IBM models.

Example model families to verify in tenant catalog:

• Granite chat/instruct models
• Granite code models

Capabilities: Chat, code

7. Perplexity (aicore-perplexity)

Perplexity AI models accessed through SAP AI Core where enabled for the tenant.

Example model families to verify in tenant catalog:

• Sonar-family web-grounded chat models
• Deep-research models with citations, where enabled

Capabilities: Chat with citations, web-grounded responses

Note: Sonar and Sonar Pro models support an output-with-citations feature in orchestration, returning source URLs alongside responses.

---

API: List Available Models

curl -X GET "$AI_API_URL/v2/lm/scenarios/foundation-models/models" \
  -H "Authorization: Bearer $AUTH_TOKEN" \
  -H "AI-Resource-Group: default" \
  -H "Content-Type: application/json"

Response Structure

{
  "count": 50,
  "resources": [
    {
      "model": "gpt-4o",
      "accessType": "Remote",
      "displayName": "GPT-4o",
      "provider": "azure-openai",
      "allowedScenarios": ["foundation-models"],
      "executableId": "azure-openai",
      "description": "OpenAI's most advanced model",
      "versions": [
        {
          "name": "2024-05-13",
          "isLatest": true,
          "capabilities": ["text-generation", "chat", "vision"],
          "contextLength": 128000,
          "inputCost": 5.0,
          "outputCost": 15.0,
          "deprecationDate": null,
          "retirementDate": null,
          "isStreamingSupported": true
        }
      ]
    }
  ]
}

Model Metadata Fields

---

Deploying a Model

Step 1: Create Configuration

curl -X POST "$AI_API_URL/v2/lm/configurations" \
  -H "Authorization: Bearer $AUTH_TOKEN" \
  -H "AI-Resource-Group: default" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "gpt4o-deployment-config",
    "executableId": "azure-openai",
    "scenarioId": "foundation-models",
    "parameterBindings": [
      {"key": "modelName", "value": "gpt-4o"},
      {"key": "modelVersion", "value": "latest"}
    ]
  }'

Step 2: Create Deployment

curl -X POST "$AI_API_URL/v2/lm/deployments" \
  -H "Authorization: Bearer $AUTH_TOKEN" \
  -H "AI-Resource-Group: default" \
  -H "Content-Type: application/json" \
  -d '{
    "configurationId": "<config-id-from-step-1>"
  }'

Step 3: Check Status

curl -X GET "$AI_API_URL/v2/lm/deployments/<deployment-id>" \
  -H "Authorization: Bearer $AUTH_TOKEN" \
  -H "AI-Resource-Group: default"

Wait for status RUNNING and note the deploymentUrl.

---

Using the Harmonized API

The harmonized API provides unified access without model-specific code.

Chat Completion

curl -X POST "$DEPLOYMENT_URL/chat/completions" \
  -H "Authorization: Bearer $AUTH_TOKEN" \
  -H "AI-Resource-Group: default" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-4o",
    "messages": [
      {"role": "system", "content": "You are a helpful assistant."},
      {"role": "user", "content": "What is SAP AI Core?"}
    ],
    "max_tokens": 1000,
    "temperature": 0.7
  }'

With Streaming

curl -X POST "$DEPLOYMENT_URL/chat/completions" \
  -H "Authorization: Bearer $AUTH_TOKEN" \
  -H "AI-Resource-Group: default" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-4o",
    "messages": [{"role": "user", "content": "Tell me a story"}],
    "stream": true
  }'

Embeddings

curl -X POST "$DEPLOYMENT_URL/embeddings" \
  -H "Authorization: Bearer $AUTH_TOKEN" \
  -H "AI-Resource-Group: default" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "text-embedding-3-large",
    "input": ["Document chunk to embed"],
    "encoding_format": "float"
  }'

---

Orchestration Deployment

For unified access to multiple models:

Create Orchestration Deployment

# Get orchestration configuration ID
curl -X GET "$AI_API_URL/v2/lm/configurations?scenarioId=orchestration" \
  -H "Authorization: Bearer $AUTH_TOKEN" \
  -H "AI-Resource-Group: default"

# Create deployment
curl -X POST "$AI_API_URL/v2/lm/deployments" \
  -H "Authorization: Bearer $AUTH_TOKEN" \
  -H "AI-Resource-Group: default" \
  -H "Content-Type: application/json" \
  -d '{
    "configurationId": "<orchestration-config-id>"
  }'

Use Orchestration API

curl -X POST "$ORCHESTRATION_URL/v2/completion" \
  -H "Authorization: Bearer $AUTH_TOKEN" \
  -H "AI-Resource-Group: default" \
  -H "Content-Type: application/json" \
  -d '{
    "config": {
      "module_configurations": {
        "llm_module_config": {
          "model_name": "gpt-4o",
          "model_version": "latest"
        },
        "templating_module_config": {
          "template": [
            {"role": "user", "content": "{{?prompt}}"}
          ]
        }
      }
    },
    "input_params": {
      "prompt": "What is machine learning?"
    }
  }'

---

Model Version Management

Auto-Upgrade Strategy

Set modelVersion to "latest" for automatic upgrades:

{
  "parameterBindings": [
    {"key": "modelName", "value": "gpt-4o"},
    {"key": "modelVersion", "value": "latest"}
  ]
}

Pinned Version Strategy

Specify exact version for stability:

{
  "parameterBindings": [
    {"key": "modelName", "value": "gpt-4o"},
    {"key": "modelVersion", "value": "2024-05-13"}
  ]
}

Manual Version Upgrade

Patch deployment with new configuration:

curl -X PATCH "$AI_API_URL/v2/lm/deployments/<deployment-id>" \
  -H "Authorization: Bearer $AUTH_TOKEN" \
  -H "AI-Resource-Group: default" \
  -H "Content-Type: application/json" \
  -d '{
    "configurationId": "<new-config-id>"
  }'

---

SAP AI Launchpad UI

Prompt Experimentation

Access: Workspaces → Generative AI Hub → Prompt Editor

Features:

• Interactive prompt testing
• Model selection and parameter tuning
• Variable placeholders
• Image inputs (select models)
• Streaming responses
• Save prompts (manager roles)

Required Roles

Prompt Types

• Question Answering: Q&A interactions
• Summarization: Extract key points
• Inferencing: Sentiment, entity extraction
• Transformations: Translation, format conversion
• Expansions: Content generation

---

Model Library

View model specifications and benchmarks in AI Launchpad:

Access: Generative AI Hub → Model Library

Information available:

• Model capabilities
• Context window sizes
• Performance benchmarks (win rates, arena scores)
• Cost per token
• Deprecation schedules

---

Rate Limits and Quotas

Refer to SAP Note 3437766 for:

• Token conversion rates per model
• Rate limits (requests/minute, tokens/minute)
• Regional availability
• Deprecation dates

Quota Increase Request

Submit support ticket:

• Component: CA-ML-AIC
• Include: tenant ID, current limits, requested limits, justification

---

Best Practices

Model Selection

Cost Optimization

1. Use smaller models for simple tasks 2. Implement caching for repeated queries 3. Set appropriate max_tokens limits 4. Use streaming for better UX without extra cost 5. Monitor token usage via AI Launchpad analytics

Reliability

1. Implement fallback configurations 2. Pin model versions in production 3. Monitor deprecation dates 4. Test before upgrading versions

---

Documentation Links

• Generative AI Hub: https://help.sap.com/docs/sap-ai-core/generative-ai/generative-ai-hub
• Supported Models: https://help.sap.com/docs/sap-ai-core/generative-ai/supported-models
• SAP Note 3437766: Token rates, limits, deprecation
• SAP Discovery Center: https://discovery-center.cloud.sap/serviceCatalog/sap-ai-core

Grounding and RAG Reference

Complete reference for SAP AI Core grounding capabilities (Retrieval-Augmented Generation).

Documentation Source: https://github.com/SAP-docs/sap-artificial-intelligence/tree/main/docs/sap-ai-core

---

Overview

Grounding integrates external, contextually relevant data into AI processes, enhancing LLM capabilities beyond general training material using vector databases.

Key Benefits

• Provide domain-specific context
• Access real-time data
• Reduce hallucinations
• Enable enterprise knowledge retrieval

---

Architecture

Indexing Pipeline

Documents → Preprocessing → Chunking → Embedding → Vector Database

1. Upload documents to supported repository 2. Pipeline preprocesses and chunks documents 3. Embedding model generates vectors 4. Vectors stored in managed vector database

Retrieval Pipeline

User Query → Embedding → Vector Search → Retrieved Chunks → LLM Context

1. User query converted to embedding 2. Vector similarity search in database 3. Relevant chunks retrieved 4. Chunks injected into LLM prompt

---

Supported Data Sources

---

Document Specifications

Supported Formats

Limits

• Maximum documents per pipeline: 2,000
• Refresh rate: Daily automatic refresh
• File size: Varies by format

---

Data Management APIs

Three primary APIs for document processing and retrieval:

Pipelines API

Creates data management pipelines that fetch documents from supported data sources.

Vector API

REST APIs for direct document ingestion and retrieval using vector embeddings.

Retrieval API

Performs similarity searches on the vector database.

API Comparison

---

Implementation Options

Option 1: Pipeline API

Automated document processing pipeline.

Create SharePoint Pipeline

curl -X POST "$AI_API_URL/v2/lm/groundingPipelines" \
  -H "Authorization: Bearer $AUTH_TOKEN" \
  -H "AI-Resource-Group: default" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "hr-policies-pipeline",
    "configuration": {
      "dataSource": {
        "type": "sharepoint",
        "configuration": {
          "siteUrl": "https://company.sharepoint.com/sites/HR",
          "folderPath": "/Documents/Policies"
        }
      },
      "secretName": "sharepoint-credentials"
    }
  }'

Create S3 Pipeline

curl -X POST "$AI_API_URL/v2/lm/groundingPipelines" \
  -H "Authorization: Bearer $AUTH_TOKEN" \
  -H "AI-Resource-Group: default" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "knowledge-base-pipeline",
    "configuration": {
      "dataSource": {
        "type": "s3",
        "configuration": {
          "bucket": "my-knowledge-base",
          "prefix": "documents/"
        }
      },
      "secretName": "s3-credentials"
    }
  }'

Create SFTP Pipeline

curl -X POST "$AI_API_URL/v2/lm/groundingPipelines" \
  -H "Authorization: Bearer $AUTH_TOKEN" \
  -H "AI-Resource-Group: default" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "docs-sftp-pipeline",
    "configuration": {
      "dataSource": {
        "type": "sftp",
        "configuration": {
          "host": "sftp.company.com",
          "port": 22,
          "path": "/documents"
        }
      },
      "secretName": "sftp-credentials"
    }
  }'

Option 2: Vector API

Direct vector upload for custom chunking/embedding.

Create Collection

curl -X POST "$AI_API_URL/v2/lm/groundingCollections" \
  -H "Authorization: Bearer $AUTH_TOKEN" \
  -H "AI-Resource-Group: default" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "custom-knowledge-base",
    "embeddingConfig": {
      "model": "text-embedding-3-small",
      "dimensions": 1536
    }
  }'

Note: Use text-embedding-3-small for 1536 dimensions or text-embedding-3-large with 3072 dimensions. Ensure model and dimensions align with OpenAI/SAP AI Core specifications.

Add Documents

curl -X POST "$AI_API_URL/v2/lm/groundingCollections/{collectionId}/documents" \
  -H "Authorization: Bearer $AUTH_TOKEN" \
  -H "AI-Resource-Group: default" \
  -H "Content-Type: application/json" \
  -d '{
    "documents": [
      {
        "id": "doc-001",
        "content": "Document chunk text...",
        "metadata": {
          "source": "policy-manual.pdf",
          "page": 5,
          "department": "HR"
        }
      },
      {
        "id": "doc-002",
        "content": "Another chunk...",
        "metadata": {
          "source": "policy-manual.pdf",
          "page": 6,
          "department": "HR"
        }
      }
    ]
  }'

Add Pre-computed Vectors

curl -X POST "$AI_API_URL/v2/lm/groundingCollections/{collectionId}/documents" \
  -H "Authorization: Bearer $AUTH_TOKEN" \
  -H "AI-Resource-Group: default" \
  -H "Content-Type: application/json" \
  -d '{
    "documents": [
      {
        "id": "doc-001",
        "content": "Document chunk text...",
        "vector": [0.123, -0.456, 0.789, ...],
        "metadata": {"source": "manual.pdf"}
      }
    ]
  }'

---

Creating Secrets

SharePoint Secret

curl -X POST "$AI_API_URL/v2/admin/secrets" \
  -H "Authorization: Bearer $AUTH_TOKEN" \
  -H "AI-Resource-Group: default" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "sharepoint-credentials",
    "data": {
      "clientId": "<azure-app-client-id>",
      "clientSecret": "<azure-app-client-secret>",
      "tenantId": "<azure-tenant-id>"
    }
  }'

S3 Secret

curl -X POST "$AI_API_URL/v2/admin/secrets" \
  -H "Authorization: Bearer $AUTH_TOKEN" \
  -H "AI-Resource-Group: default" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "s3-credentials",
    "data": {
      "AWS_ACCESS_KEY_ID": "<access-key>",
      "AWS_SECRET_ACCESS_KEY": "<secret-key>",
      "AWS_REGION": "us-east-1"
    }
  }'

SFTP Secret

curl -X POST "$AI_API_URL/v2/admin/secrets" \
  -H "Authorization: Bearer $AUTH_TOKEN" \
  -H "AI-Resource-Group: default" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "sftp-credentials",
    "data": {
      "username": "<username>",
      "password": "<password>"
    }
  }'

---

Using Grounding in Orchestration

Basic Grounding Configuration

{
  "config": {
    "module_configurations": {
      "grounding_module_config": {
        "grounding_service": "document_grounding_service",
        "grounding_service_configuration": {
          "grounding_input_parameters": ["user_query"],
          "grounding_output_parameter": "context",
          "filters": [
            {
              "id": "<pipeline-id>",
              "search_configuration": {
                "max_chunk_count": 5
              }
            }
          ]
        }
      },
      "templating_module_config": {
        "template": [
          {
            "role": "system",
            "content": "Answer based on the following context:\n\n{{$context}}\n\nIf the answer is not in the context, say you don't know."
          },
          {
            "role": "user",
            "content": "{{?user_query}}"
          }
        ]
      },
      "llm_module_config": {
        "model_name": "gpt-4o",
        "model_version": "latest"
      }
    }
  },
  "input_params": {
    "user_query": "What is the vacation policy?"
  }
}

Grounding with Metadata Filters

{
  "grounding_module_config": {
    "grounding_service": "document_grounding_service",
    "grounding_service_configuration": {
      "grounding_input_parameters": ["user_query"],
      "grounding_output_parameter": "context",
      "filters": [
        {
          "id": "<pipeline-id>",
          "data_repositories": ["<specific-repo-id>"],
          "document_metadata": [
            {
              "key": "department",
              "value": "HR"
            },
            {
              "key": "document_type",
              "value": "policy"
            }
          ],
          "search_configuration": {
            "max_chunk_count": 10,
            "max_document_count": 5,
            "similarity_threshold": 0.7
          }
        }
      ]
    }
  }
}

Multiple Pipeline Sources

{
  "grounding_module_config": {
    "grounding_service": "document_grounding_service",
    "grounding_service_configuration": {
      "grounding_input_parameters": ["user_query"],
      "grounding_output_parameter": "context",
      "filters": [
        {
          "id": "<hr-pipeline-id>",
          "search_configuration": {"max_chunk_count": 3}
        },
        {
          "id": "<it-pipeline-id>",
          "search_configuration": {"max_chunk_count": 3}
        },
        {
          "id": "<finance-pipeline-id>",
          "search_configuration": {"max_chunk_count": 3}
        }
      ]
    }
  }
}

---

Search Configuration Options

---

Managing Pipelines

List Pipelines

curl -X GET "$AI_API_URL/v2/lm/groundingPipelines" \
  -H "Authorization: Bearer $AUTH_TOKEN" \
  -H "AI-Resource-Group: default"

Get Pipeline Status

curl -X GET "$AI_API_URL/v2/lm/groundingPipelines/{pipelineId}" \
  -H "Authorization: Bearer $AUTH_TOKEN" \
  -H "AI-Resource-Group: default"

Pipeline Statuses:

• PENDING: Initializing
• INDEXING: Processing documents
• READY: Available for queries
• FAILED: Error occurred

Delete Pipeline

curl -X DELETE "$AI_API_URL/v2/lm/groundingPipelines/{pipelineId}" \
  -H "Authorization: Bearer $AUTH_TOKEN" \
  -H "AI-Resource-Group: default"

---

Best Practices

Document Preparation

1. Clean content: Remove irrelevant headers, footers, boilerplate 2. Consistent formatting: Use clear headings and structure 3. Metadata tagging: Add useful metadata for filtering 4. Regular updates: Keep documents current

Chunking Strategy

1. Semantic chunks: Break at logical boundaries (sections, paragraphs) 2. Appropriate size: 200-500 tokens per chunk typically works well 3. Overlap: Consider 10-20% overlap between chunks 4. Context preservation: Include section headers in chunks

Query Optimization

1. Clear questions: Rephrase vague queries 2. Keyword inclusion: Include relevant technical terms 3. Context addition: Add domain context to queries

Retrieval Tuning

---

Troubleshooting

No Results Returned

1. Check pipeline status is READY 2. Verify documents were indexed successfully 3. Lower similarity threshold 4. Increase max_chunk_count 5. Check metadata filters match documents

Irrelevant Results

1. Increase similarity threshold 2. Add metadata filters 3. Review document chunking 4. Check embedding model matches query style

Performance Issues

1. Reduce max_chunk_count 2. Add specific metadata filters 3. Use multiple smaller pipelines 4. Consider pagination for large result sets

---

Documentation Links

• Grounding Overview: https://github.com/SAP-docs/sap-artificial-intelligence/blob/main/docs/sap-ai-core/grounding-035c455.md
• Pipeline API: https://github.com/SAP-docs/sap-artificial-intelligence/blob/main/docs/sap-ai-core/create-a-document-grounding-pipeline-using-the-pipelines-api-0a13e1c.md
• SharePoint Pipeline: https://github.com/SAP-docs/sap-artificial-intelligence/blob/main/docs/sap-ai-core/create-a-pipeline-with-microsoft-sharepoint-4b8d58c.md
• S3 Pipeline: https://github.com/SAP-docs/sap-artificial-intelligence/blob/main/docs/sap-ai-core/create-a-pipeline-with-aws-s3-7f97adf.md

ML Operations Reference

Complete reference for SAP AI Core ML training and operations.

Documentation Source: https://github.com/SAP-docs/sap-artificial-intelligence/tree/main/docs/sap-ai-core

---

Overview

SAP AI Core uses Argo Workflows for training pipelines, supporting batch jobs for model preprocessing, training, and inference.

Key Components

---

Workflow Engine

Argo Workflows

SAP AI Core uses Argo Workflows (container-native workflow engine) supporting:

• Direct Acyclic Graph (DAG) structures
• Parallel step execution
• Container-based steps
• Data ingestion and preprocessing
• Model training and batch inference

Limitation: Not optimized for time-critical tasks due to scheduling overhead.

---

Prerequisites

1. Object Store Secret (Required)

Create a secret named default for training output artifacts:

curl -X POST "$AI_API_URL/v2/admin/objectStoreSecrets" \
  -H "Authorization: Bearer $AUTH_TOKEN" \
  -H "AI-Resource-Group: default" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "default",
    "type": "S3",
    "pathPrefix": "my-bucket/training-output",
    "data": {
      "AWS_ACCESS_KEY_ID": "<access-key>",
      "AWS_SECRET_ACCESS_KEY": "<secret-key>"
    }
  }'

Note: Without a default secret, training pipelines will fail.

2. Docker Registry Secret

For custom training images:

curl -X POST "$AI_API_URL/v2/admin/dockerRegistrySecrets" \
  -H "Authorization: Bearer $AUTH_TOKEN" \
  -H "AI-Resource-Group: default" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "docker-registry",
    "data": {
      ".dockerconfigjson": "<base64-encoded-docker-config>"
    }
  }'

3. Git Repository

Sync workflow templates from Git:

curl -X POST "$AI_API_URL/v2/admin/repositories" \
  -H "Authorization: Bearer $AUTH_TOKEN" \
  -H "AI-Resource-Group: default" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "training-repo",
    "url": "https://github.com/org/training-workflows",
    "username": "<git-user>",
    "password": "<git-token>"
  }'

---

Workflow Template

Basic Structure

apiVersion: ai.sap.com/v1alpha1
kind: WorkflowTemplate
metadata:
  name: text-classifier-training
  annotations:
    scenarios.ai.sap.com/description: "Train text classification model"
    scenarios.ai.sap.com/name: "text-classifier"
    executables.ai.sap.com/description: "Training executable"
    executables.ai.sap.com/name: "text-classifier-train"
    artifacts.ai.sap.com/training-data.kind: "dataset"
    artifacts.ai.sap.com/trained-model.kind: "model"
  labels:
    scenarios.ai.sap.com/id: "text-classifier"
    executables.ai.sap.com/id: "text-classifier-train"
    ai.sap.com/version: "1.0.0"
spec:
  imagePullSecrets:
    - name: docker-registry
  entrypoint: main
  arguments:
    parameters:
      - name: learning_rate
        default: "0.001"
      - name: epochs
        default: "10"
    artifacts:
      - name: training-data
        path: /data/input
        archive:
          none: {}
  templates:
    - name: main
      steps:
        - - name: preprocess
            template: preprocess-data
        - - name: train
            template: train-model
        - - name: evaluate
            template: evaluate-model

    - name: preprocess-data
      container:
        image: my-registry/preprocessing:latest
        command: ["python", "preprocess.py"]
        args: ["--input", "/data/input", "--output", "/data/processed"]

    - name: train-model
      container:
        image: my-registry/training:latest
        command: ["python", "train.py"]
        args:
          - "--data=/data/processed"
          - "--lr={{workflow.parameters.learning_rate}}"
          - "--epochs={{workflow.parameters.epochs}}"
          - "--output=/data/model"
      outputs:
        artifacts:
          - name: trained-model
            path: /data/model
            globalName: trained-model
            archive:
              none: {}

    - name: evaluate-model
      container:
        image: my-registry/evaluation:latest
        command: ["python", "evaluate.py"]
        args: ["--model", "/data/model"]

Annotations Reference

---

Artifacts

Types

Register Input Artifact

curl -X POST "$AI_API_URL/v2/lm/artifacts" \
  -H "Authorization: Bearer $AUTH_TOKEN" \
  -H "AI-Resource-Group: default" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "training-dataset-v1",
    "kind": "dataset",
    "url": "ai://default/datasets/training-v1",
    "scenarioId": "text-classifier",
    "description": "Training dataset version 1"
  }'

URL Format

• ai://default/<path> - Uses default object store secret
• ai://<secret-name>/<path> - Uses named object store secret

List Artifacts

curl -X GET "$AI_API_URL/v2/lm/artifacts?scenarioId=text-classifier" \
  -H "Authorization: Bearer $AUTH_TOKEN" \
  -H "AI-Resource-Group: default"

---

Configurations

Create Training Configuration

curl -X POST "$AI_API_URL/v2/lm/configurations" \
  -H "Authorization: Bearer $AUTH_TOKEN" \
  -H "AI-Resource-Group: default" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "text-classifier-config-v1",
    "executableId": "text-classifier-train",
    "scenarioId": "text-classifier",
    "parameterBindings": [
      {"key": "learning_rate", "value": "0.001"},
      {"key": "epochs", "value": "20"},
      {"key": "batch_size", "value": "32"}
    ],
    "inputArtifactBindings": [
      {"key": "training-data", "artifactId": "<dataset-artifact-id>"}
    ]
  }'

---

Executions

Create Execution

curl -X POST "$AI_API_URL/v2/lm/executions" \
  -H "Authorization: Bearer $AUTH_TOKEN" \
  -H "AI-Resource-Group: default" \
  -H "Content-Type: application/json" \
  -d '{
    "configurationId": "<configuration-id>"
  }'

Execution Statuses

Check Execution Status

curl -X GET "$AI_API_URL/v2/lm/executions/<execution-id>" \
  -H "Authorization: Bearer $AUTH_TOKEN" \
  -H "AI-Resource-Group: default"

Get Execution Logs

curl -X GET "$AI_API_URL/v2/lm/executions/<execution-id>/logs" \
  -H "Authorization: Bearer $AUTH_TOKEN" \
  -H "AI-Resource-Group: default"

Stop Execution

curl -X PATCH "$AI_API_URL/v2/lm/executions/<execution-id>" \
  -H "Authorization: Bearer $AUTH_TOKEN" \
  -H "AI-Resource-Group: default" \
  -H "Content-Type: application/json" \
  -d '{"targetStatus": "STOPPED"}'

---

Metrics

Write Metrics from Training

In your training code:

import requests
import os

def log_metrics(metrics: dict, step: int):
    """Log metrics to SAP AI Core."""
    api_url = os.environ.get("AICORE_API_URL")
    token = os.environ.get("AICORE_AUTH_TOKEN")
    execution_id = os.environ.get("AICORE_EXECUTION_ID")

    response = requests.post(
        f"{api_url}/v2/lm/executions/{execution_id}/metrics",
        headers={
            "Authorization": f"Bearer {token}",
            "Content-Type": "application/json"
        },
        json={
            "metrics": [
                {"name": name, "value": value, "step": step}
                for name, value in metrics.items()
            ]
        }
    )

# Usage in training loop
for epoch in range(epochs):
    train_loss = train_epoch()
    val_loss = validate()
    log_metrics({
        "train_loss": train_loss,
        "val_loss": val_loss,
        "accuracy": accuracy
    }, step=epoch)

Read Metrics

curl -X GET "$AI_API_URL/v2/lm/executions/<execution-id>/metrics" \
  -H "Authorization: Bearer $AUTH_TOKEN" \
  -H "AI-Resource-Group: default"

---

Training Schedules

Create Schedule

curl -X POST "$AI_API_URL/v2/lm/executionSchedules" \
  -H "Authorization: Bearer $AUTH_TOKEN" \
  -H "AI-Resource-Group: default" \
  -H "Content-Type: application/json" \
  -d '{
    "configurationId": "<configuration-id>",
    "cron": "0 0 * * 0",
    "start": "2024-01-01T00:00:00Z",
    "end": "2024-12-31T23:59:59Z"
  }'

Cron Expression Format

SAP AI Core uses 5-field cron expressions with 3-letter day-of-week names:

┌───────── minute (0-59)
│ ┌─────── hour (0-23)
│ │ ┌───── day of month (1-31)
│ │ │ ┌─── month (1-12)
│ │ │ │ ┌─ day of week (mon, tue, wed, thu, fri, sat, sun)
│ │ │ │ │
* * * * *

Examples:

• 0 0 * * * - Daily at midnight
• 0 0 * * sun - Weekly on Sunday
• 0 0 * * fri - Weekly on Friday
• 0 0 1 * * - Monthly on 1st
• 0 */6 * * * - Every 6 hours

Note: Using * * * * * treats the schedule as "Run Always" (continuous check), which differs from standard cron behavior. Minimum interval for pipeline schedules is 1 hour.

List Schedules

curl -X GET "$AI_API_URL/v2/lm/executionSchedules" \
  -H "Authorization: Bearer $AUTH_TOKEN" \
  -H "AI-Resource-Group: default"

Delete Schedule

curl -X DELETE "$AI_API_URL/v2/lm/executionSchedules/<schedule-id>" \
  -H "Authorization: Bearer $AUTH_TOKEN" \
  -H "AI-Resource-Group: default"

---

SAP AI Launchpad

ML Operations App

Access: Workspaces → ML Operations

Features:

• View scenarios and executables
• Create/manage configurations
• Run/monitor executions
• View training metrics
• Manage artifacts
• Create schedules

Required Roles

Comparing Runs

1. Navigate to ML Operations → Executions 2. Select multiple executions 3. Click "Compare" 4. View side-by-side metrics and parameters

---

Best Practices

Workflow Design

1. Modular steps: Break workflow into reusable templates 2. Parameterization: Use parameters for hyperparameters 3. Artifact management: Define clear input/output artifacts 4. Error handling: Include retry logic for flaky operations

Resource Management

1. Appropriate sizing: Match container resources to workload 2. GPU allocation: Request GPUs only when needed 3. Storage: Use object store for large datasets 4. Cleanup: Delete old executions and artifacts

Monitoring

1. Log metrics: Track loss, accuracy, etc. during training 2. Check logs: Review execution logs for errors 3. Compare runs: Analyze different hyperparameter settings 4. Set alerts: Monitor for failed executions

---

Troubleshooting

Execution Failed

1. Check execution logs: GET /v2/lm/executions/{id}/logs 2. Verify object store secret exists and is named default 3. Check Docker image is accessible 4. Verify artifact paths are correct 5. Check resource quota not exceeded

Artifacts Not Found

1. Verify artifact URL format: ai://default/<path> 2. Check object store secret permissions 3. Verify file exists in object store 4. Check artifact registered in correct scenario

Schedule Not Running

1. Verify schedule is active (not paused) 2. Check cron expression is valid 3. Verify start/end dates bracket current time 4. Check configuration still exists

---

Documentation Links

• Training Overview: https://github.com/SAP-docs/sap-artificial-intelligence/blob/main/docs/sap-ai-core/train-your-model-a9ceb06.md
• ML Operations (Launchpad): https://github.com/SAP-docs/sap-artificial-intelligence/blob/main/docs/sap-ai-launchpad/ml-operations-df78271.md
• Schedules: https://github.com/SAP-docs/sap-artificial-intelligence/blob/main/docs/sap-ai-core/create-a-training-schedule-bd409a9.md
• Metrics: https://github.com/SAP-docs/sap-artificial-intelligence/blob/main/docs/sap-ai-core/view-the-metric-resource-for-an-execution-d85dd44.md

Model Providers Reference

Complete reference for SAP AI Core model providers and available models.

Documentation Source: SAP Help Portal - SAP AI Core

Latest Models: SAP Note 3437766

---

Overview

SAP AI Core provides access to models from multiple providers via the Generative AI Hub. All models are accessed through a unified API, allowing easy switching between providers.

Catalog rule: This reference describes provider families and common configuration shapes. Exact model IDs, versions, context windows, pricing, regions, and deprecation dates must be verified in the target tenant through SAP AI Launchpad Model Library, GET /v2/lm/scenarios/foundation-models/models, and SAP Note 3437766 before implementation.

---

Provider Summary

---

1. Azure OpenAI

Executable ID: azure-openai Access Type: Remote (Azure-hosted)

Example Model Families

Deprecated/retiring patterns: older GPT-4, GPT-4 Turbo, GPT-4-32k, and GPT-3.5-era deployments should be checked against SAP Note 3437766 and migrated before retirement dates shown in the tenant catalog.

Configuration Example

{
  "name": "azure-gpt4o-config",
  "executableId": "azure-openai",
  "scenarioId": "foundation-models",
  "parameterBindings": [
    {"key": "modelName", "value": "gpt-4o"},
    {"key": "modelVersion", "value": "2024-05-13"}
  ]
}

---

2. SAP-Hosted Open Source

Executable ID: aicore-opensource Access Type: Local (SAP-hosted)

Example Model Families

Configuration Example

{
  "name": "llama-config",
  "executableId": "aicore-opensource",
  "scenarioId": "foundation-models",
  "parameterBindings": [
    {"key": "modelName", "value": "meta--llama-3.1-70b-instruct"},
    {"key": "modelVersion", "value": "latest"}
  ]
}

---

3. Google Vertex AI

Executable ID: gcp-vertexai Access Type: Remote (Google Cloud)

Example Model Families

Deprecated/retiring patterns: older Gemini and PaLM-era deployments should be checked against SAP Note 3437766 and migrated before retirement dates shown in the tenant catalog.

Configuration Example

{
  "name": "gemini-config",
  "executableId": "gcp-vertexai",
  "scenarioId": "foundation-models",
  "parameterBindings": [
    {"key": "modelName", "value": "gemini-1.5-pro"},
    {"key": "modelVersion", "value": "latest"}
  ]
}

---

4. AWS Bedrock

Executable ID: aws-bedrock Access Type: Remote (AWS)

Example Model Families

Configuration Example

{
  "name": "claude-config",
  "executableId": "aws-bedrock",
  "scenarioId": "foundation-models",
  "parameterBindings": [
    {"key": "modelName", "value": "anthropic--claude-3-5-sonnet"},
    {"key": "modelVersion", "value": "latest"}
  ]
}

---

5. Mistral AI

Executable ID: aicore-mistralai Access Type: Local (SAP-hosted)

Models

Configuration Example

{
  "name": "mistral-config",
  "executableId": "aicore-mistralai",
  "scenarioId": "foundation-models",
  "parameterBindings": [
    {"key": "modelName", "value": "mistralai--mistral-large"},
    {"key": "modelVersion", "value": "latest"}
  ]
}

---

6. IBM

Executable ID: aicore-ibm Access Type: Local (SAP-hosted)

Granite Models

Configuration Example

{
  "name": "granite-config",
  "executableId": "aicore-ibm",
  "scenarioId": "foundation-models",
  "parameterBindings": [
    {"key": "modelName", "value": "ibm--granite-13b-chat"},
    {"key": "modelVersion", "value": "latest"}
  ]
}

---

7. Perplexity

Executable ID: aicore-perplexity Access Type: Remote (Perplexity-hosted) Availability: tenant-dependent; verify in SAP AI Core catalog.

Example Model Families

Unique Features:

• Returns citation URLs alongside responses
• Web-grounded responses for up-to-date information
• Supports output-with-citations in orchestration workflows

Configuration Example

{
  "name": "perplexity-config",
  "executableId": "aicore-perplexity",
  "scenarioId": "foundation-models",
  "parameterBindings": [
    {"key": "modelName", "value": "perplexity--sonar"},
    {"key": "modelVersion", "value": "latest"}
  ]
}

---

Model Selection Guide

By Use Case

By Budget

By Capability

---

Model Version Management

Version Strategies

Checking Available Versions

curl -X GET "$AI_API_URL/v2/lm/scenarios/foundation-models/models" \
  -H "Authorization: Bearer $AUTH_TOKEN" \
  -H "AI-Resource-Group: default" | \
  jq '.resources[] | select(.model == "gpt-4o") | .versions'

Handling Deprecation

1. Monitor deprecationDate in model metadata 2. Plan migration before retirementDate 3. Test new version in staging 4. Update configuration with new version 5. Patch existing deployments

---

Pricing Considerations

Pricing varies by:

• Model complexity (larger = more expensive)
• Input vs output tokens (output often 2-3x input cost)
• Provider region
• Access type (Remote vs Local)

Reference: SAP Note 3437766 for current token rates.

Cost Optimization

1. Right-size models: Use smaller models for simple tasks 2. Batch requests: Combine multiple queries when possible 3. Cache responses: Store and reuse common query results 4. Limit tokens: Set appropriate max_tokens limits 5. Use streaming: No additional cost, better UX

---

Rate Limits

Rate limits vary by:

• Service plan tier
• Model provider
• Specific model

Default limits (vary by configuration):

• Requests per minute: 60-600
• Tokens per minute: 40K-400K

Handling Rate Limits

import time
from requests.exceptions import HTTPError

def call_with_retry(func, max_retries=3):
    for attempt in range(max_retries):
        try:
            return func()
        except HTTPError as e:
            if e.response.status_code == 429:
                wait_time = 2 ** attempt
                time.sleep(wait_time)
            else:
                raise
    raise Exception("Max retries exceeded")

---

Documentation Links

• Supported Models: https://help.sap.com/docs/sap-ai-core/generative-ai/supported-models
• Generative AI Hub: https://help.sap.com/docs/sap-ai-core/generative-ai/generative-ai-hub
• SAP Note 3437766: Token rates, limits, deprecation dates
• SAP Discovery Center: https://discovery-center.cloud.sap/serviceCatalog/sap-ai-core