Rust-backed Client Migration Guide

In an effort to ensure all clients have the same set of capabilities we havemigrated the Python and Node clients onto a common Rust base library. In Python,both the synchronous and asynchronous clients are based on this implementation.In Node, the new client is available as@lancedb/lancedb, which replacesthe existingvectordb package.

This guide describes the differences between the two Node APIs and will hopefully assist usersthat would like to migrate to the new API.

TypeScript/JavaScript

For JS/TS users, we offer a brand new SDK@lancedb/lancedb

We tried to keep the API as similar as possible to the previous version, but there are a few small changes. Here are the most important ones:

Creating Tables

CreateTableOptions.writeOptions.writeMode has been replaced withCreateTableOptions.mode

db.createTable(tableName,data,{writeMode:lancedb.WriteMode.Overwrite});

db.createTable(tableName,data,{mode:"overwrite"})

Changes to Table APIs

PreviouslyTable.schema was a property. Now it is an async method.

Creating Indices

TheTable.createIndex method is now used for creating both vector indicesand scalar indices. It currently requires a column name to be specified (thecolumn to index). Vector index defaults are now smarter and scale better withthe size of the data.

awaittbl.createIndex({column:"vector",// defaulttype:"ivf_pq",num_partitions:2,num_sub_vectors:2,});

awaittable.createIndex("vector",{config:lancedb.Index.ivfPq({numPartitions:2,numSubVectors:2,}),});

Embedding Functions

The embedding API has been completely reworked, and it now more closely resembles the Python API, including the newembedding registry:

constembeddingFunction=newlancedb.OpenAIEmbeddingFunction('text',API_KEY)constdata=[{id:1,text:'Black T-Shirt',price:10},{id:2,text:'Leather Jacket',price:50}]consttable=awaitdb.createTable('vectors',data,embeddingFunction)

import*aslancedbfrom"@lancedb/lancedb";import*asarrowfrom"apache-arrow";import{LanceSchema,getRegistry}from"@lancedb/lancedb/embedding";constfunc=getRegistry().get("openai").create({apiKey:API_KEY});constdata=[{id:1,text:'Black T-Shirt',price:10},{id:2,text:'Leather Jacket',price:50}]consttable=awaitdb.createTable('vectors',data,{embeddingFunction:{sourceColumn:"text",function:func,}})

You can also use a schema driven approach, which parallels the Pydantic integration in our Python SDK:

constfunc=getRegistry().get("openai").create({apiKey:API_KEY});constdata=[{id:1,text:'Black T-Shirt',price:10},{id:2,text:'Leather Jacket',price:50}]constschema=LanceSchema({id:newarrow.Int32(),text:func.sourceField(newarrow.Utf8()),price:newarrow.Float64(),vector:func.vectorField()})consttable=awaitdb.createTable('vectors',data,{schema})