sap-cloud-sdk-ai-python

# SAP Cloud SDK for AI (Python) > **Package rename**: The PyPI package `generative-ai-hub-sdk` is **deprecated** (v4.12.4 is the last release). > Its successor is **`sap-ai-sdk-gen`** (currently v6.10.0 per public PyPI registry evidence from 2026-06-15). Code and tutorials referencing > `generative-ai-hub-sdk` should migrate to `sap-ai-sdk-gen`; the import name remains `gen_ai_hub`. The official Python SDK for SAP Generative AI Hub and Orchestration Service. It wraps the native SDKs of model providers (OpenAI, Amazon Bedrock, Google GenAI) and offers a harmonised LangChain integration and a full Orchestration client — all routed through SAP AI Core with unified authentication. Package freshness is registry-verified; AI Core runtime behavior and exact model availability still require target-tenant validation. ## Related Skills - **sap-ai-core**: Platform setup, deployments, resource groups, and model management in SAP AI Core - **sap-cloud-sdk-ai**: JavaScript/TypeScript and Java equivalents of this SDK - **sap-hana-ml**: HANA-side machine learning in Python - **sap-dependency-security**: Pip dependency hygiene and upgrade patterns ### Related external skills If your task involves working inside Databricks (notebooks, Unity Catalog, Spark, SAP Databricks in SAP Business Data Cloud), consider installing the [Databricks agent skills plugin](https://github.com/databricks/databricks-agent-skills). Ask whether you would like help installing it — never install unprompted. ## When to Use This Skill Use this skill when: - Building Python applications that call LLMs through SAP AI Core / Generative AI Hub - Using the `gen_ai_hub` Python package (installed as `sap-ai-sdk-gen`) - Integrating OpenAI, Amazon Bedrock, or Google GenAI models via SAP's proxy - Implementing LangChain chains with SAP AI Core as the backend - Using the Orchestration Service from Python (templating, filtering, masking, grounding) - Migrating code from the deprecated `generative-ai-hub-sdk` to `sap-ai-sdk-gen` - Generating embeddings through SAP AI Core - Working with SAP RPT-1 (Relational Pretrained Transformer) for tabular predictions ## Table of Contents - [Quick Start](#quick-start) - [Installation](#installation) - [Authentication](#authentication) - [Available Modules](#available-modules) - [Supported Models](#supported-models) - [Core Features](#core-features) - [Bundled Resources](#bundled-resources) ## Quick Start ### Native OpenAI Chat Completion ```python from gen_ai_hub.proxy.native.openai import chat messages = [ {"role": "system", "content": "You are a helpful assistant."}, {"role": "user", "content": "What is SAP BTP?"} ] response = chat.completions.create( model_name="gpt-4o-mini", messages=messages ) print(response.choices[0].message.content) ``` ### Orchestration Service ```python from gen_ai_hub.orchestration_v2 import ( OrchestrationConfig, OrchestrationService, ModuleConfig, PromptTemplatingModuleConfig, Template, UserMessage, LLMModelDetails ) config = OrchestrationConfig( modules=ModuleConfig( prompt_templating=PromptTemplatingModuleConfig( prompt=Template( template=[UserMessage(role="user", content="{{?question}}")] ), model=LLMModelDetails(name="gpt-4o-mini") ) ) ) service = OrchestrationService(config=config) response = service.run(placeholder_values={"question": "What is SAP?"}) print(response.final_result.choices[0].message.content) ``` ## Installation ```bash # All providers + LangChain support pip install "sap-ai-sdk-gen[all]" # Default (OpenAI only, no LangChain) pip install sap-ai-sdk-gen # Specific providers (without LangChain) pip install "sap-ai-sdk-gen[google, amazon]" ``` ## Authentication The SDK reads credentials via `AICoreV2Client.from_env()`, which resolves credentials in this order: 1. **Keyword arguments** passed to `GenAIHubProxyClient(...)` 2. **Environment variables** — `AICORE_CLIENT_ID`, `AICORE_CLIENT_SECRET`, `AICORE_AUTH_URL`, `AICORE_BASE_URL`, `AICORE_RESOURCE_GROUP` 3. **Config file** — `$AICORE_HOME/config.json` (or path set by `AICORE_CONFIG`); use `AICORE_PROFILE` to select a named profile 4. **VCAP_SERVICES** — automatic on Cloud Foundry/Kyma when the AI Core service is bound ### Local Development (Environment Variables) ```bash export AICORE_CLIENT_ID="sb-..." export AICORE_CLIENT_SECRET="..." export AICORE_AUTH_URL="https://<tenant>.authentication.sap.hana.ondemand.com/oauth/token" export AICORE_BASE_URL="https://api.ai.prod.eu-central-1.aws.ml.hana.ondemand.com/v2" export AICORE_RESOURCE_GROUP="default" ``` ### Config File Profile ```bash # ~/.aicore/config.json { "AICORE_CLIENT_ID": "sb-...", "AICORE_CLIENT_SECRET": "...", "AICORE_AUTH_URL": "https://<tenant>.authentication.sap.hana.ondemand.com/oauth/token", "AICORE_BASE_URL": "https://api.ai.prod.eu-central-1.aws.ml.hana.ondemand.com/v2", "AICORE_RESOURCE_GROUP": "default" } ``` For detailed auth setup and troubleshooting, see `references/getting-started-auth.md`. ## Available Modules | Module | Import Path | Purpose | |--------|-------------|---------| | Proxy (native clients) | `gen_ai_hub.proxy.native.*` | Direct model access per provider | | LangChain integration | `gen_ai_hub.proxy.langchain` | `init_llm`, `init_embedding_model`, `ChatOpenAI`, etc. | | Orchestration | `gen_ai_hub.orchestration_v2` | Templating, filtering, masking, grounding | | Document Grounding | `gen_ai_hub.document_grounding` | Pipeline, Vector, Retrieval APIs | | Prompt Registry | `gen_ai_hub.prompt_registry` | Template management and config storage | | Evaluations | `gen_ai_hub.evaluations` | Model evaluation runs and metrics | | SAP RPT-1 | `gen_ai_hub.proxy.native.sap` | Tabular prediction (classification, regression) | ### Native Clients by Provider | Provider | Import | Key Classes | |----------|--------|-------------| | OpenAI | `gen_ai_hub.proxy.native.openai` | `OpenAI`, `completions`, `chat`, `embeddings`, `responses` | | Amazon Bedrock | `gen_ai_hub.proxy.native.amazon` | `Session`, `ClientWrapper` | | Google GenAI | `gen_ai_hub.proxy.native.google_genai` | `Client` | | SAP RPT-1 | `gen_ai_hub.proxy.native.sap` | `RPTClient`, `RPTRequest` | ## Supported Models The Generative AI Hub catalog includes models from multiple providers. Check [SAP's model catalog](https://help.sap.com/docs/sap-ai-core/generative-ai-hub/available-models) and the target tenant catalog for the authoritative model IDs. Example families: | Provider | Example Families | |----------|------------------| | OpenAI | GPT-family chat, multimodal, reasoning, and embedding models | | Anthropic (via Bedrock) | Claude-family models | | Amazon | Nova/Titan-family models | | Google | Gemini-family models | | Mistral | Mistral-family models | | SAP | RPT-family tabular prediction models where enabled | ## Core Features ### Chat Completion with OpenAI Client ```python from gen_ai_hub.proxy.native.openai import OpenAI client = OpenAI() response = client.chat.completions.create( model="gpt-4o-mini", messages=[{"role": "user", "content": "Explain CAP in one paragraph."}] ) print(response.choices[0].message.content) ``` ### Streaming ```python from gen_ai_hub.proxy.native.openai import OpenAI client = OpenAI() stream = client.chat.completions.create( model="gpt-4o-mini", messages=[{"role": "user", "content": "Explain SAP CAP."}], stream=True ) for chunk in stream: if chunk.choices[0].delta.content: print(chunk.choices[0].delta.content, end="") ``` ### Embeddings ```python from gen_ai_hub.proxy.native.openai import embeddings response = embeddings.create( input="Every decoding is another encoding.", model_name="text-embedding-3-small" ) print(response.data[0].embedding) ``` ### LangChain Integration ```python from gen_ai_hub.proxy.langchain import init_llm, init_embedding_model llm = init_llm("gpt-4o-mini", max_tokens=300) result = llm.invoke("What is SAP BTP?") print(result.content) embeddings = init_embedding_model("text-embedding-3-small") vector = embeddings.embed_query("SAP Business Technology Platform") ``` ### Content Filtering (via Orchestration) ```python from gen_ai_hub.orchestration_v2 import ( OrchestrationConfig, OrchestrationService, ModuleConfig, PromptTemplatingModuleConfig, Template, UserMessage, LLMModelDetails, FilteringModuleConfig, InputFiltering, OutputFiltering, AzureContentSafetyInput, AzureContentSafetyOutput, AzureThreshold ) config = OrchestrationConfig( modules=ModuleConfig( prompt_templating=PromptTemplatingModuleConfig( prompt=Template(template=[UserMessage(role="user", content="{{?question}}")]), model=LLMModelDetails(name="gpt-4o-mini") ), filtering=FilteringModuleConfig( input=InputFiltering(filters=[ AzureContentSafetyInput(hate=AzureThreshold.ALLOW_SAFE, violence=AzureThreshold.ALLOW_SAFE) ]), output=OutputFiltering(filters=[ AzureContentSafetyOutput(hate=AzureThreshold.ALLOW_SAFE, violence=AzureThreshold.ALLOW_SAFE) ]) ) ) ) service = OrchestrationService(config=config) response = service.run(placeholder_values={"question": "Explain SAP."}) ``` ### Data Masking (via Orchestration) ```python from gen_ai_hub.orchestration_v2 import ( OrchestrationConfig, OrchestrationService, ModuleConfig, PromptTemplatingModuleConfig, Template, UserMessage, LLMModelDetails, MaskingModuleConfig, MaskingProviderConfig, DPIStandardEntity, MaskingMethod, DataMaskingProviderName ) config = OrchestrationConfig( modules=ModuleConfig( prompt_templating=PromptTemplatingModuleConfig( prompt=Template(template=[UserMessage(role="user", content="{{?text}}")]), model=LLMModelDetails(name="gpt-4o-mini") ), masking=MaskingModuleConfig( masking_providers=[ MaskingProviderConfig( type=DataMaskingProviderName.SAP_DATA_PRIVACY_INTEGRATION, method=MaskingMethod.ANONYMIZATION, entities=[ DPIStandardEntity(type="profile-email"), DPIStandardEntity(type="profile-person") ] ) ] ) ) ) service = OrchestrationService(config=config) response = service.run(placeholder_values={"text": "Contact john@example.com for details."}) ``` ### Document Grounding (via Orchestration) ```python from gen_ai_hub.orchestration_v2 import ( OrchestrationConfig, OrchestrationService, ModuleConfig, PromptTemplatingModuleConfig, Template, UserMessage, LLMModelDetails, GroundingModuleConfig, DocumentGroundingConfig, DocumentGroundingFilter, DocumentGroundingPlaceholders, GroundingSearchConfig, DataRepositoryType, GroundingType ) config = OrchestrationConfig( modules=ModuleConfig( prompt_templating=PromptTemplatingModuleConfig( prompt=Template(template=[UserMessage(role="user", content="{{?question}}")]), model=LLMModelDetails(name="gpt-4o-mini") ), grounding=GroundingModuleConfig( type=GroundingType.DOCUMENT_GROUNDING_SERVICE, config=DocumentGroundingConfig( placeholders=DocumentGroundingPlaceholders( input=["{{?question}}"], output="{{?context}}" ), filters=[ DocumentGroundingFilter( id="my-vector-repo-id", data_repository_type=DataRepositoryType.VECTOR, search_config=GroundingSearchConfig(max_chunk_count=5) ) ] ) ) ) ) service = OrchestrationService(config=config) response = service.run(placeholder_values={"question": "What is the refund policy?"}) ``` ## Common Errors | Error | Cause | Solution | |-------|-------|----------| | `No credentials found in any source` | Missing AI Core service key/env vars | Set all `AICORE_*` environment variables or create a config file profile | | `No deployment found` | Model not deployed in AI Core | Deploy the model in your resource group, or use `deployment_id` directly |