@reaatech/hybrid-rag-embedding

Status: Pre-1.0 — APIs may change in minor versions. Pin to a specific version in production.

Provider-agnostic embedding generation for hybrid RAG systems. Currently supports OpenAI embeddings with extension points for Vertex AI and local models.

Installation

npm install @reaatech/hybrid-rag-embedding
# or
pnpm add @reaatech/hybrid-rag-embedding

Feature Overview

• Provider abstraction — single interface across OpenAI, Vertex AI, and local models
• Batch processing — configurable batch size with automatic rate limiting
• Cost tracking — per-request cost calculation based on model pricing
• Dimension lookup — static method to get the embedding dimension for known models
• OpenAI integration — full text-embedding-3-small and text-embedding-3-large support

Quick Start

import { EmbeddingService } from '@reaatech/hybrid-rag-embedding';

const embedder = new EmbeddingService({
  provider: 'openai',
  model: 'text-embedding-3-small',
  apiKey: process.env.OPENAI_API_KEY,
  batchSize: 100,
  rateLimit: 3500, // requests per minute
});

// Single text
const { embedding, tokens, cost } = await embedder.embed('Hello world');
console.log(`Vector dimension: ${embedding.length}, Cost: $${cost}`);

// Batch
const texts = ['Document one...', 'Document two...', 'Document three...'];
const results = await embedder.embedBatch(texts);
const totalCost = results.reduce((sum, r) => sum + r.cost, 0);

API Reference

EmbeddingService

Constructor

new EmbeddingService(config: EmbeddingConfig)

EmbeddingConfig

| Property | Type | Default | Description | |----------|------|---------|-------------| | provider | 'openai' \| 'vertex' \| 'local' | (required) | Embedding provider | | model | string | (required) | Model name (e.g. text-embedding-3-small) | | apiKey | string | — | API key for cloud providers | | dimension | number | — | Embedding dimension (auto-detected for known models) | | batchSize | number | 100 | Max texts per batch request | | rateLimit | number | — | Requests per minute throttle |

Methods

| Method | Returns | Description | |--------|---------|-------------| | embed(text) | Promise<EmbeddingResult> | Generate embedding for a single text | | embedBatch(texts) | Promise<EmbeddingResult[]> | Generate embeddings for multiple texts with batching and rate limiting |

Static Methods

| Method | Returns | Description | |--------|---------|-------------| | getDimension(model) | number | Get the dimension for a known model (1536 for text-embedding-3-small, 3072 for text-embedding-3-large) |

EmbeddingResult

| Property | Type | Description | |----------|------|-------------| | embedding | number[] | The embedding vector | | tokens | number | Number of tokens consumed | | cost | number | Cost in USD |

Cost Calculation

Pricing is built-in for known models:

| Model | Cost per 1M tokens | |-------|-------------------| | text-embedding-3-small | $0.02 | | text-embedding-3-large | $0.13 |

Provider Extensibility

The Vertex AI and local providers are extension points. To add a new provider:

class CustomEmbedder extends EmbeddingService {
  private async embedCustom(text: string): Promise<EmbeddingResult> {
    // Your implementation here
    return { embedding: [/* ... */], tokens: 0, cost: 0 };
  }
}

Related Packages

• @reaatech/hybrid-rag — Core types
• @reaatech/hybrid-rag-qdrant — Qdrant vector DB (consumes embeddings)

License

MIT