Skip to content

Vector Database by Relevance AI. Fast and simple with support for chaining.

License

Notifications You must be signed in to change notification settings

RelevanceAI/relevance-js-sdk

Repository files navigation

relevance-js-sdk

Install with npm using:

npm i @relevanceai/dataset

Features

  • Node and Browser support
  • Typescript definitions for almost all relevanceai.com apis
  • Insert millions of documents with one function call
  • Our SearchBuilder makes searching, filtering, and aggregating your data simple

Getting started

Get started by creating an account in cloud.relevanceai.com - select the Vector Database onboarding option. Once set up you can fetch your API key and use the below snippet.

import {Client,QueryBuilder} from "@relevanceai/dataset";

const discovery = new Client({
  project: '',
  api_key: '',
  endpoint: ''
});
const dataset = discovery.dataset('1000-movies');

const movies = [{ title: 'Lord of the Rings: The Fellowship of the Ring', grenre: 'action', budget: 100 }, ...]
await dataset.insertDocuments(movies, [{ model_name: 'text-embedding-ada-002', field: 'title' }]);

const {results} = await dataset.search(QueryBuilder().vector('title_vector_', { query: 'LOTR', model: 'text-embeddings-ada-002' }));

Set up your credentials

Option 1 - Use environment variables

First, set environment variables in your shell before you run your code.

set RELEVANCE_PROJECT to your project name.

set RELEVANCE_API_KEY to your api key. for more information, view the docs here: Authorization docs

Heres a template to copy and paste in for linux environments:

export RELEVANCE_PROJECT=#########
export RELEVANCE_API_KEY=#########

The SDK will use these variables when making api calls. You can then initialise your client like this:

import {Client} from "@relevanceai/dataset";
const client = new Client({});

Option 2 - Passing them in code.

import {Client} from "@relevanceai/dataset";
const client = new Client({
  project:'########',
  api_key:'########',
});

Examples

You can import builders and type definitions like this

import {QueryBuilder,Client,BulkInsertOutput} from "@relevanceai/dataset";

Insert millions of items with one function call

const discovery = new Client({ ... });
const dataset = discovery.dataset('tshirts-prod');
 // Here we create some demo data. Replace this with your real data
const fakeVector = [];
for (let i = 0; i < 768; i++) fakeVector.push(1);
const tshirtsData = [];
for (let i = 0; i < 10000; i++) {
  tshirtsData.push({_id:`tshirt-${i}1`,color:'red',price:i/1000,'title-fake_vector_':fakeVector});
  tshirtsData.push({_id:`tshirt-${i}2`,color:'blue',price:i/1000});
  tshirtsData.push({_id:`tshirt-${i}3`,color:'orange',price:i/1000});
}
const res = await dataset.insertDocuments(tshirtsData,{batchSize:10000});

insertDocuments will output:

{"inserted":30000,"failed_documents":[]}

Text Search and Vector Search

const builder = QueryBuilder();
builder.query('red').text().vector('title-fake_vector_',0.5).minimumRelevance(0.1);
// .text() searches all fields. alternatively, use .text(field1).text(field2)... to search specific fields
const searchResults = await dataset.search(builder);

Filter and retrieve items

const filters = QueryBuilder();
filters.match('color',['blue','red']).range('price',{lessThan:50});
const filteredItems = await dataset.search(filters);

search will output:

{
  results: [
    {
      color: 'red',
      price: 0,
      insert_date_: '2021-11-16T03:14:28.509Z',
      _id: 'tshirt-01',
      _relevance: 0
    }
    ...
  ],
  resultsSize: 10200,
  aggregations: {},
  aggregates: {},
  aggregateStats: {}
}
## Call raw api methods directly
```javascript
const discovery = new Client({ ... });
const dataset = discovery.dataset('tshirts-prod');
const {body} = await dataset.apiClient.FastSearch({filters:[{match:{key:'_id',value:`tshirt-01`}}]});
expect((body.results[0] as any).color).toBe('red')