sap-cloud-sdk-ai

name	sap-cloud-sdk-ai
description	SAP Cloud SDK for AI — generative AI hub SDK, LLM orchestration, prompt templating, model selection, orchestration service for multi-step AI workflows, document grounding, retrieval augmented generation (RAG). Use when building AI agents with SAP Cloud SDK, calling LLMs via SAP BTP GenAI Hub, or implementing RAG patterns with SAP data.
trigger	{"keywords":["ai","sdk","generative","llm","orchestration","rag","prompt","btp","genai","models"],"intent":"Build AI agents and orchestrate LLMs on SAP BTP using the Cloud SDK for AI, including prompt templating, RAG, and model selection."}

SAP Cloud SDK for AI

Generative AI development on SAP BTP — LLM orchestration, prompt management, RAG.

Python SDK

from gen_ai_hub.orchestration import OrchestrationService

# Orchestration with templating and grounding
service = OrchestrationService(
    api_url="https://api.ai.core.prod.<region>.aws.cloud.sap/v2",
    client_id="...", client_secret="..."
)

# Define a prompt template
template = '''
System: You are an SAP procurement expert. Answer using only provided context.

Context: {{?grounding_context}}

User: {{?user_input}}
'''

result = service.invoke(
    template=template,
    grounding={"type": "document", "source": "sap-procurement-docs"},
    model={"name": "gpt-4"},
    input={"user_input": "How do I create a purchase order?"}
)
print(result.content)

Document Grounding

# Upload SAP documents for RAG
service.upload_document(
    name="sap-procurement-docs",
    content=open("procurement_guide.pdf", "rb"),
    metadata={"category": "SAP_MM", "language": "EN"}
)

# Query with grounding
response = service.query_with_grounding(
    question="What BAPI creates a purchase order?",
    document_names=["sap-procurement-docs"],
    top_k=3
)

Model Selection

Model	Provider	Best For
GPT-4 / GPT-4o	OpenAI	Complex reasoning, code generation
Codex Opus / Sonnet	Anthropic	Analysis, structured output
Llama 3	Meta (self-hosted)	Cost-sensitive, offline scenarios
Gemini Pro	Google	Multimodal, long context

Multi-Step Orchestration

# Chain: classify → route → generate → validate
orchestrator = OrchestrationService()

steps = [
    {"name": "classify", "template": classify_template, "model": "gpt-4o-mini"},
    {"name": "generate", "template": generate_template, "model": "gpt-4"},
    {"name": "validate", "template": validate_template, "model": "gpt-4o-mini"}
]

result = orchestrator.run_pipeline(steps=steps, input={"question": user_input})

Gotchas

• GenAI Hub requires AI Core entitlement on BTP subaccount
• Rate limits vary by model provider (OpenAI vs Anthropic vs Meta)
• Document grounding documents must be in supported format (PDF, TXT, MD)
• Always validate LLM output — hallucination risk with SAP technical content