Concept card

SAP Cloud SDK for AI & CAP LLM plugin

As of 2026-09-23

Edited by Cédric-Olivier Mâry, sole editor and publication director — 27 years in the SAP ecosystem. How these pages are produced, and where automation is used.

What is SAP Cloud SDK for AI & CAP LLM plugin?

SAP Cloud SDK for AI (JavaScript/TypeScript, Java, Python) is SAP's official client for SAP AI Core, the Generative AI Hub and the orchestration service, with LangChain and Spring AI integrations. The older CAP LLM plugin was archived in June 2026 — new CAP projects combine the SDK with CAP's native AI features instead.

What it is

SAP Cloud SDK for AI is the official software development kit for SAP AI Core, the Generative AI Hub and its orchestration service. It gives application code a typed, authenticated client for chat completion, embeddings, document grounding, prompt management and AI Core lifecycle APIs, so developers do not hand-craft REST calls, token handling or deployment-URL lookups. It exists in three languages:

JavaScript/TypeScript — npm packages under @sap-ai-sdk/*: orchestration, foundation-models, langchain, document-grounding, prompt-registry and ai-api (AI Core scenarios, executions, deployments). Version 2.16.0 was published on 18 September 2026.
Java — Maven group com.sap.ai.sdk (modules include orchestration, foundationmodels, document-grounding, prompt-registry, core); latest release 1.25.0 in early September 2026. Requires Java 17 or higher (21 LTS recommended); integrates with Spring AI (≥ 1.0.0), Spring Boot ≥ 3.0 and CAP Java ≥ 3.0.0. Model classes marked @Beta may change in minor releases.
Python — package sap-ai-sdk-gen (7.4.0, September 2026, Python ≥ 3.9), which wraps native provider SDKs (OpenAI, Amazon, Google), LangChain and the orchestration service. It replaces the older generative-ai-hub-sdk, which PyPI now labels deprecated.

Why it matters

SAP Cloud SDK for AI is the official, maintained client for AI Core and the Generative AI Hub in JavaScript, Java and Python — the default building block of every pro-code SAP AI extension.
The CAP LLM plugin was archived on 16 June 2026; material that still recommends it is outdated and new projects should use the SDK plus CAP's native AI features.
The orchestration service gives a harmonised API across model providers plus data masking, filtering and grounding — the enterprise controls a direct model call lacks.

Key points

Three SDKs: JavaScript/TypeScript (@sap-ai-sdk/*), Java (com.sap.ai.sdk), Python (sap-ai-sdk-gen).
JS packages: orchestration, foundation-models, langchain, document-grounding, prompt-registry, ai-api; version 2.16.0 on 18 Sep 2026.
Java: JDK 17+ (21 LTS recommended), Spring AI ≥ 1.0.0, Spring Boot ≥ 3.0, CAP Java ≥ 3.0.0; release 1.25.0 in Sep 2026.
Python: sap-ai-sdk-gen 7.4.0 (Python ≥ 3.9) replaces the deprecated generative-ai-hub-sdk.
Orchestration service = templating, data masking, content filtering, grounding, translation, fallback around one harmonised model API.
LangChain: @sap-ai-sdk/langchain OrchestrationClient (JS), Spring AI integration (Java), sap-ai-sdk-gen[all] (Python).
CAP LLM plugin: last release 1.5.1 (Sep 2025), repository archived 16 Jun 2026.
CAP native AI replaces it: Vector type + vector_embedding() in HANA Cloud, @cap-js/mcp, @cap-js/agents.

Terms used on this page

SAP Cloud SDK for AI: SAP's official SDK family (JavaScript/TypeScript, Java, Python) for SAP AI Core, the Generative AI Hub and the orchestration service.
Orchestration service: Generative AI Hub service that wraps a model call with templating, data masking, content filtering, grounding, translation and fallback behind one harmonised API.
Foundation-models client: SDK package for calling a specific provider's model API directly through AI Core, without orchestration modules.
Resource group: AI Core scoping unit that isolates deployments, configurations and secrets per tenant or project.
CAP LLM plugin: Former CAP Node.js plugin for anonymisation, embeddings, similarity search and RAG; repository archived on 16 June 2026.
vector_embedding(): SAP HANA Cloud function, usable from CAP calculated elements, that computes embeddings inside the database.
AICORE_SERVICE_KEY: Environment variable the SDK reads for local development to authenticate against AI Core; production uses bindings or destinations.

Sources

Full card available to members. What the full card adds: the full decision framework · the common pitfalls and their fix · the cheat sheet · the code blocks · the facts worth quoting.

Open in the app →