Aicore Sdk Js

Rules

1. If no deployment URL is known, check the SAP AI Core console or contact your admin to list or create one before generating code.
2. Always include aicore configure credential guidance in generated code comments; link to Connecting to AI Core for setup details.
3. JS/TS lifecycle management (create/list/delete deployments) is not available in this skill — use the SAP AI Core console or contact your admin for deployment management.
4. Target @sap-ai-sdk v2 (orchestration 2.14+). The preferred invocation pattern is passing messages directly to chatCompletion() / stream(). Only use promptTemplating.prompt.template with {{?placeholder}} values when the user explicitly wants a server-side reusable template. Never emit the v1 shape llm: { model_name } / templating: { template } — it does not compile against v2.
5. Never emit Python code snippets — this skill is JS/TS only. If the user asks for Python, inform them that no Python skill is available in this repo and point them to https://sap.github.io/ai-sdk/ for Python SDK guidance.
6. Gather these details before generating code — ask only for what the user hasn't provided:
   • Framework: @sap-ai-sdk/* (recommended) / @sap-ai-sdk/langchain (LangChain) / OpenAI npm package via LiteLLM proxy?
   • Use case: chat / streaming / embeddings?
   • Runtime: Node.js / Bun / browser?
   • Model + Deployment ID: check the SAP AI Core console for existing deployments
   • Model-switching intent (ask only when the user specifies an OpenAI model like gpt-4o, gpt-4.1, etc.): do they plan to switch models later (e.g. to Claude or Gemini)? If yes → recommend @sap-ai-sdk/orchestration for portability. If they want to stay on OpenAI → use @sap-ai-sdk/openai (SapOpenAi.createClient) as the simpler native OpenAI SDK alternative; no LiteLLM proxy needed.

Sources & Fallback

Use the best source available, in this order. Never block on a source that isn't present.

1. Baked-in reference (always available). references/SDK_JS.md and assets/*.ts are the compile-checked v2 floor. They are correct for the common paths (chat, streaming, LangChain, embeddings, proxy) and require no network or MCP. Use them by default.
2. Context7 MCP (use if available). If the context7 query-docs tool is registered in this session, prefer it for the freshest API, version-specific details, or any feature the baked-in reference marks as fast-moving (tool calling, data masking, document grounding, prompt registry). Library ID: /websites/sap_github_io_ai-sdk_js. If the tool is not available, silently fall back to the baked-in reference — do not ask the user to install it.
3. Web docs (last resort). For anything neither source covers, point to llms.txt or fetch the online docs.

Do not fabricate API shapes from memory — this SDK's v1 and v2 APIs differ, and guessing reproduces stale v1 code. When unsure, use source 2 if present, else source 3.

Asset Map

Install Options

bashCopy code
# @sap-ai-sdk (recommended — native SAP SDK)
npm install @sap-ai-sdk/orchestration @sap-ai-sdk/foundation-models

# Native OpenAI SDK surface on AI Core (no LiteLLM proxy needed)
npm install @sap-ai-sdk/openai openai zod

# LangChain integration (no LiteLLM proxy needed)
npm install @sap-ai-sdk/langchain @langchain/core

# RAG, prompt templates, batch jobs, tabular prediction
npm install @sap-ai-sdk/document-grounding
npm install @sap-ai-sdk/prompt-registry
npm install @sap-ai-sdk/llm-batch @sap-ai-sdk/ai-api @sap-ai-sdk/foundation-models
npm install @sap-ai-sdk/rpt

# With bun
bun add @sap-ai-sdk/orchestration @sap-ai-sdk/foundation-models

# OpenAI npm package via LiteLLM proxy (only if you need a running proxy; prefer @sap-ai-sdk/openai)
npm install openai

Quick Start (@sap-ai-sdk/orchestration)

Pass messages directly — this is the default pattern. Only use promptTemplating.prompt.template with {{?placeholder}} values when the user explicitly wants a reusable server-side template.

typescriptCopy code
import { OrchestrationClient } from "@sap-ai-sdk/orchestration";

const client = new OrchestrationClient({
  promptTemplating: {
    model: {
      name: "anthropic--claude-4.5-haiku",
      params: { temperature: 0.7 },
    },
  },
});

const response = await client.chatCompletion({
  messages: [
    { role: "system", content: "You are a helpful assistant." },
    { role: "user", content: "Hello!" },
  ],
});
console.log(response.getContent());

Run with: npx tsx index.ts or bun run index.ts

Handoffs

• Want Java code instead? → invoke aicore-sdk-java

When to Load Reference Files