How do I load text from a SQLite table and correctly chunk it with LangChain?

I'm taking my first steps with LangChain and have some text stored in a SQLite table that I want to load and chunk using the RecursiveCharacterTextSplitter.

I tried using better-sqlite3 to query my table and then passing each row as a document to the splitter. However, I ran into an error:

JAVASCRIPTCopied
import { RecursiveCharacterTextSplitter } from "@langchain/textsplitters";
const splitter = new RecursiveCharacterTextSplitter({
    chunkSize: 1000, 
    chunkOverlap: 200
});

// given CREATE TABLE t (id INTEGER PRIMARY KEY, fulltext TEXT);
const rows = db.prepare('SELECT id, fulltext FROM t').all();

for (const row of rows) {
   const docs = [ { metadata: row.id, pageContent: row.fulltext } ];
   const chunks = await splitter.splitDocuments(docs);
}

// Error
file:///Users/punkish/Projects/zai/node_modules/@langchain/textsplitters/dist/text_splitter.js:102
                const loc = _metadatas[i].loc && typeof _metadatas[i].loc === "object"
                                          ^

TypeError: Cannot read properties of undefined (reading 'loc')

The error seems to be related to the metadata passed to the splitter. I would like to know how to correctly pass the text selected from the table to the text splitter so that it processes the chunks as expected.