@mcp-abap-adt/sap-aicore-embedder

SAP AI Core embedding provider (IEmbedderBatch) for @mcp-abap-adt/llm-agent.

Generates text embeddings via SAP AI Core embedding model deployments using the @sap-ai-sdk/orchestration client.

Installation

npm install @mcp-abap-adt/sap-aicore-embedder

Usage

import { SapAiCoreEmbedder } from '@mcp-abap-adt/sap-aicore-embedder';

const embedder = new SapAiCoreEmbedder({
  model: 'text-embedding-3-small',
  resourceGroup: 'default', // optional
});

// Embed a single text
const result = await embedder.embed('Hello, world!');
console.log(result.vector);

// Batch embed multiple texts
const results = await embedder.embedBatch(['Text 1', 'Text 2', 'Text 3']);
console.log(results.map(r => r.vector));

Configuration

interface SapAiCoreEmbedderConfig {
  model: string;
  resourceGroup?: string;                             // default: 'default'
  scenario?: 'foundation-models' | 'orchestration';   // default: 'orchestration'
  credentials?: FoundationModelsCredentials;          // foundation-models only; falls back to AICORE_SERVICE_KEY
}

Orchestration (default)

The default scenario — matches v11.0.0 behavior. Use this when the embedding model is deployed under the orchestration scenario. Authentication is handled automatically by the SAP AI SDK orchestration client using AICORE_SERVICE_KEY.

import { SapAiCoreEmbedder } from '@mcp-abap-adt/sap-aicore-embedder';

const embedder = new SapAiCoreEmbedder({ model: 'text-embedding-3-small' });

Existing v11.0.0 consumers require no config changes.

Foundation-models

Opt-in path for tenants where embedding models (such as gemini-embedding or text-embedding-3-small) are deployed under the foundation-models scenario rather than the orchestration scenario. The embedder calls the AI Core REST inference API directly.

const embedder = new SapAiCoreEmbedder({
  model: 'gemini-embedding',
  scenario: 'foundation-models',
});
// Auth: process.env.AICORE_SERVICE_KEY (client_credentials flow)

Authentication

For the orchestration scenario (default), authentication is handled automatically by the SAP AI SDK orchestration client using AICORE_SERVICE_KEY.

For the foundation-models scenario, authentication uses the AICORE_SERVICE_KEY environment variable directly (client credentials flow). You can also pass credentials explicitly via the credentials option.

License

MIT