ballerinax/weaviate Ballerina library

Overview

This is a generated connector for the Weaviate Vector Search Engine API OpenAPI specification. Weaviate is an open-source vector search engine, which allows storing data objects and vector embeddings from the ML models, including the LLMs offered by OpenAI, Hugging Face, and Cohere. Weaviate provides a powerful GraphQL API for querying the embeddings while looking at the similarity and can scale seamlessly into billions of data objects.

Prerequisites

Before using this connector in your Ballerina application, complete the following:

1. Create a Weaviate Cluster using the Weaviate Cloud Service or deploy using Docker/Kubernetes.
2. Obtain the OIDC Authentication key.

Quick start

To use the Weaviate connector in your Ballerina application, update the .bal file as follows:

Step 1: Import the connector

First, import the ballerinax/weaviate module into the Ballerina project.

import ballerinax/weaviate;

Step 2: Create a new connector instance

Create and initialize a weaviate:Client with your Service URL and the obtained Authentication key.

weaviate:Client weaviateClient = check new ({
    auth: {
        token: "sk-XXXXXXXXX"
    }
}, serviceURL: "https://weaviate-server:port/v1");

Step 3: Invoke the connector operation

1. Now, you can use the operations available within the connector. Following is an example of inserting new objects to the Weaviate vector storage as a batch operation.

   string className = "DocClass"; // weaviate class name
   string[] text; // list of text
   float[][] embeddings; // list of embbedings for the texts
   int len = text.length();
   
   // Creates the batch of Weaviate objects.
   weaviate:Object[] objArr = [];
   foreach int i in 0...len {
       objArr.push(
           {
               'class: className,
               vector: embeddings[i],
               properties: {
                   "docs": text[i]
               }
           });
   }
   
   weaviate:ObjectsGetResponse[] responseArray = check weaviateClient->/batch/objects.post({
       objects: objArr
   });

   Once the new records are inserted, you can query the Weaviate vector storage using the Weaviate GraphQL API, similar to the example below.

   float[] embeddings;  // This is the embedding for the text being searched for similar content.
   
   string graphQLQuery =  string`{
                               Get {
                                   DocStore (
                                   nearVector: {
                                       vector: ${embeddings.toString()}
                                       }
                                       limit: 5
                                   ){
                                   docs
                                   _additional {
                                       certainty,
                                       id
                                       }
                                   }
                               }
                           }`;
   
   weaviate:GraphQLResponse|error results = check weaviateClient->/graphql.post({
       query: graphQLQuery
   });

2. Use the bal run command to compile and run the Ballerina program.