zeroentropy-ai-provider

Vercel AI SDK provider for ZeroEntropy — supports text embeddings and reranking.

Installation

npm install zeroentropy-ai-provider

Setup

Get an API key from the ZeroEntropy dashboard and set it as an environment variable:

export ZEROENTROPY_API_KEY=your_api_key

Or pass it directly when creating the provider.

Usage

Embeddings

import { createZeroEntropy } from 'zeroentropy-ai-provider';
import { embedMany } from 'ai';

const ze = createZeroEntropy(); // reads ZEROENTROPY_API_KEY from env

const { embeddings } = await embedMany({
  model: ze.textEmbeddingModel('zembed-1'),
  values: ['ZeroEntropy is a retrieval pipeline', 'Hello world'],
  experimental_providerMetadata: {
    zeroentropy: {
      inputType: 'document', // 'query' | 'document'
    },
  },
});

Reranking

import { createZeroEntropy } from 'zeroentropy-ai-provider';
import { rerank } from 'ai';

const ze = createZeroEntropy();

const { ranking } = await rerank({
  model: ze.reranking('zerank-2'),
  query: 'Where is the Eiffel Tower?',
  values: [
    'The Eiffel Tower is located in Paris.',
    'The Amazon rainforest spans several countries.',
    'Paris is the capital of France.',
  ],
});

// ranking is sorted by descending relevance
// [{ index: 0, relevanceScore: 0.98 }, ...]

Provider options

const ze = createZeroEntropy({
  apiKey: 'your_api_key',      // default: process.env.ZEROENTROPY_API_KEY
  baseURL: 'https://...',      // override API base URL
  headers: { 'X-Custom': '1' },
});

Embedding options

Option	Type	Default	Description
inputType	'query' | 'document'	'document'	Optimizes the embedding for retrieval queries or documents
dimensions	number	2560	Output dimensions: 2560, 1280, 640, 320, 160, 80, 40
latency	'fast' | 'slow'	auto	'fast' for subsecond; 'slow' for higher throughput

Reranking options

Option	Type	Default	Description
latency	'fast' | 'slow'	auto	'fast' for subsecond; 'slow' for >10s higher throughput

Models

Model	Type
zembed-1	Text embeddings (up to 2560 dims)
zerank-2	Reranking

License

Apache 2.0