in-browser-vector-db

Features

• Supports Binary vector.
• Promise based implementation.
• Supports Web Worker.

Installation

  npm i in-browser-vector-db

Quick Start

For float32(fp32) vectors.

import { Connection, getUniqueInteger } from "in-browser-vector-db";
const connection = new Connection();
const dbConnection = await connection.openDb(dbName);
const vectorBlockConnection = await dbConnection.openVectorBlock(vectorBlockName);
await vectorBlockConnection.configureVectorBlock({ vectorDimension: 384, vectorDType: 'float32' });
const insertmanyResult = await vectorBlockConnection.operations.insertMany({ indices: [getUniqueInteger(),getUniqueInteger(),getUniqueInteger()], texts: ["what is earth?","what is web?","what is vector db"], vectors: [[0.01...],[0.01...],[0.01...]], metadataArray: [{name:"test0",age:30,hobby:["dancing"]},{name:"test1",age:40,hobby:["running"]},{name:"test2",age:50,hobby:["cooking"]}] });
const searchResult = await vectorBlockConnection.operations.search({ queryVector: [0.001...], topK: 6, vectorDistance: 'cosine', where:{ name: { $eq: "test1" }, age: { $lte: 50 }, hobby: { $nin: "dancing" } }});

For bool(uint8) vectors.

import { Connection, convertFloatToBinary, getUniqueInteger } from "in-browser-vector-db";
const connection = new Connection();
const dbConnection = await connection.openDb(dbName);
const vectorBlockConnection = await dbConnection.openVectorBlock(vectorBlockName);
await vectorBlockConnection.configureVectorBlock({ vectorDimension: 384, vectorDType: 'bool' });
const binaryVectors = convertFloatToBinary([[0.001....],[0.001....],[0.001....]]);
const insertmanyResult = await vectorBlockConnection.operations.insertMany({ indices: [getUniqueInteger(),getUniqueInteger(),getUniqueInteger()], texts: ["what is earth?","what is web?","what is vector db"], vectors: binaryVectors, metadataArray: [{name:"test0",age:30,hobby:["dancing"]},{name:"test1",age:40,hobby:["running"]},{name:"test2",age:50,hobby:["cooking"]}] });
const searchResult = await vectorBlockConnection.operations.search({ queryVector: [0.001...], topK: 6, vectorDistance: 'normHamming', where:{ name: { $eq: "test1" }, age: { $lte: 50 }, hobby: { $nin: "dancing" } }});

Details

• The configuration process of vectorblock is a one time process and the applied configurtaion cannot be modified.
• The inserted vector should be same of same data type and dimension as specified in the vectorblock configuration(configureVectorBlock()).
• Available dTypes ('float32' -> fp32 ,'bool' -> uint8).
• Available vector distance ('cosine','l2','hamming','normHamming').
• Available filter ('$eq','$ne','$gt','$lt','$gte','$lte','$in','$nin').

Documentation

Starting the connection

import { Connection } from "in-browser-vector-db";
const connection = new Connection();

Opening the database.

const dbConnection = await connection.openDb("dbTest");

Opening the vectorblock.

// case-1(when no vectorBlock is opened)
const vectorBlockConnection = await dbConnection.openVectorBlock("vbTest1");

// case-2(when already a vectorBlock is opened)

// first close the opened vectorBlock.
dbConnection.closeVectorBlock();

// then open the vectorBlock.
const vectorBlockConnection = await dbConnection.openVectorBlock("vbTest1");

Configure the vectorblock.

await vectorBlockConnection.configureVectorBlock({ vectorDimension: 768, vectorDType: 'float32' });

Insert the entry.

const insertResult = await vectorBlockConnection.operations.insert({ index: 0, text: "hello test.", vector: [0.000001 ,....], metadata: {name: "test",age: 30,hobby:["dancing","running"]} });

Insert many entries.

const insertManyResult = await vectorBlockConnection.operations.insertMany({ indices:[0], texts: ["hello test."], vectors: [[0.000001 ,....]], metadataArray: [{name: "test"}] });

Update the entry.

const updateResult = await vectorBlockConnection.operations.update(index, { text: "what about you?", vector: [0.001...],metadata:{name:"test00"} });

Update many entries.

const updateManyResult = await vectorBlockConnection.operations.updateMany(indices, { texts: ["what about you?","How are you?"], vectors: [[0.001...],[0.001...]],metadataArray:[{name:"test00"},{name:"test11"}] });

Get the entry by id.

const entryAtIndexZero = await vectorBlockConnection.operations.getByIndex(0);

Get the entries by ids.

const entries = await vectorBlockConnection.operations.getByIndices([0,1,2]);

Delete the entry by id.

const deleteResult = await vectorBlockConnection.operations.deleteByIndex(0);

Delete the entries by ids.

const deleteResults = await vectorBlockConnection.operations.deleteByIndices([0,1,2]);

Delete all entries.

const deleteAllResult = await vectorBlockConnection.operations.deleteAll();

Search the similar entries.

const serachResult = await search({queryVector: [0.001...], vectorDistance: 'cosine',topK: 5,where:{ name: { $eq: "test" }, age: { $lte: 50 }, hobby: { $in: "dancing" } }});

Close the opened vectorblock.

const closeResult =  dbConnection.closeVectorBlock();

Delete the vectorblock.

const deleteResult =  await dbConnection.deleteVectorBlock("vbTest1")

Delete the database .

 // case-1(when vectorBlock is opened)

 const connection = new Connection();
 const dbConnection = await connection.openDb('test');
 const vectorBlockConnection = await dbConnection.openVectorBlock('vbTest1');
 // first close the open vectorBlock.
 console.log(dbConnection.closeVectorBlock());
 // then delete the db.
 console.log(await connection.deleteDb('test'));

 // case-2(when no vectorBlock is opened)

 const connection = new Connection();
 const dbConnection = await connection.openDb('test');
 console.log(await connection.deleteDb('test'));