Built-in vector database

Relevance AI has built-in datasets that are NoSQL-based documents and support storing vector embeddings and vector search.

Install the SDK via:

npm install --save @relevanceai/dataset

​

What is a dataset?

A dataset is a collection of documents that are stored in a NoSQL database. Each document has a unique ID and a set of fields. Each field can be a string, number, date, or a vector. A vector is a list of numbers that represent a point in a multi-dimensional space.

You can use this dataset like you would a Table in SQL or Collection in Mongo. It has CRUD with support for endpoints like:

• Get documents
• Get documents where
• Insert documents
• Update documents
• Update documents where
• Delete documents
• Delete documents where
• Search

Every document in the dataset has an _id string field that acts as a unique identifier.

​

Support for vector embeddings

Each dataset can store one or more vector fields that have a _vector_ suffix with up to 2048 dimensions. Relevance AI’s API can automatically generate the embeddings upon insert and during search by specifying the model.

const dataset = client.dataset('1000-movies');

const movies = [
  { title: 'Lord of the Rings: The Fellowship of the Ring', genre: 'action', budget: 100 }, ...
];

await dataset.insertDocuments(movies, [
  { model_name: 'text-embedding-ada-002', field: 'title' }
]);

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

const { results } = await dataset.search(query);