Embedding & Reranking Models

Embedding models turn text into numeric vectors that capture meaning. Reranking models score how relevant a set of documents is to a query. Together they power semantic search and retrieval-augmented generation (RAG).

When to Use

• Embeddings: semantic search, clustering, deduplication, classification, and the retrieval step of a RAG pipeline
• Reranking: reordering an initial set of retrieved documents so the most relevant ones rank highest, improving precision before passing context to a language model

Embedding models return vectors, not text. Use a language model to generate the final answer from retrieved context.

AI SDK Method

Embedding models are accessed with embeddingModel() and used with the AI SDK's embed and embedMany:

import { createOLLM } from '@orgn/gateway';
import { embed, embedMany } from 'ai';

const ollm = createOLLM({ apiKey: process.env.OLLM_API_KEY });

// Single vector
const { embedding } = await embed({
  model: ollm.embeddingModel('near_qwen3_embedding_0_6b'),
  value: 'OLLM routes confidential LLM traffic.',
});

// Batch
const { embeddings } = await embedMany({
  model: ollm.embeddingModel('vercel_text_embedding_3_small'),
  values: [
    'Confidential computing protects data in use.',
    'TEEs use hardware-level encryption.',
  ],
});

Discover available embedding models with ollm.listModels({ outputModality: 'embedding' }).

TEE Catalog

Embedding and reranking models running in Trusted Execution Environments, on NEAR and Phala infrastructure with Intel TDX + NVIDIA H100 confidential compute.

ZDR Catalog

Embedding and reranking models running on Vercel's AI infrastructure with zero data retention provider agreements.

Embedding Models

Reranking Models