[Bug] The expected type comes from property 'pineconeIndex' which is declared here on type 'PineconeStoreParams'

[kirandash]

pineconeIndex type error. The issue does not occur with v3.0.3

Error information

Type 'import("/Users/kirandash/workspace/bgwebagency/ai-web-apps/node_modules/@pinecone-database/pinecone/dist/data/index").Index<import("/Users/kirandash/workspace/bgwebagency/ai-web-apps/node_modules/@pinecone-database/pinecone/dist/data/vectors/types").RecordMetadata>' is not assignable to type 'import("/Users/kirandash/workspace/bgwebagency/ai-web-apps/node_modules/@langchain/pinecone/node_modules/@pinecone-database/pinecone/dist/data/index").Index<import("/Users/kirandash/workspace/bgwebagency/ai-web-apps/node_modules/@langchain/pinecone/node_modules/@pinecone-database/pinecone/dist/data/types").RecordMet...'.
  Types have separate declarations of a private property 'config'.

Steps to reproduce the issue locally

• install the following packages

    "@langchain/community": "^0.3.6",
    "@langchain/openai": "^0.3.11",
    "@langchain/pinecone": "^0.1.1",
    "@langchain/redis": "^0.1.0",
    "@pinecone-database/pinecone": "^4.0.0",

My code:

    const vectorStore = await PineconeStore.fromExistingIndex(embeddings, {
      pineconeIndex,
      maxConcurrency: 5,
    });

[aulorbe]

Hey @kirandash , could you share your fromExistingIndex function and how you're instantiating PineconeStore as well, please? Thank you!

Without having investigated this issue quite yet, there is a high likelihood this issue stems from Langchain's Pinecone integration, rather than the Pinecone native TS client.

[kirandash]

Hey @aulorbe,

Certainly! Here’s how I’m using the fromExistingIndex function and instantiating PineconeStore:

// src/app/api/document-query/route.ts

import { pineconeIndex } from "@/lib/pinecone";
import { OpenAI, OpenAIEmbeddings } from "@langchain/openai";
import { PineconeStore } from "@langchain/pinecone";
import { VectorDBQAChain } from "langchain/chains";
import { NextResponse } from "next/server";

export async function POST(req: Request) {
  const { prompt } = await req.json();

  if (!prompt) {
    return NextResponse.json({ error: "Prompt is required" }, { status: 400 });
  }

  try {
    const model = new OpenAI();

    if (!model) {
      throw new Error("Failed to initialize OpenAI model");
    }

    const embeddings = new OpenAIEmbeddings({
      model: "text-embedding-3-small",
    });

    const vectorStore = await PineconeStore.fromExistingIndex(embeddings, {
      pineconeIndex,
      maxConcurrency: 5,
    });

    const chain = VectorDBQAChain.fromLLM(model, vectorStore, {
      k: 1,
      returnSourceDocuments: true,
    });

    const response = await chain.call({
      query: prompt,
    });

    return NextResponse.json({ result: response });
  } catch (error) {
    console.error(error);
    return NextResponse.json({ error: "An error occurred. Please try again" });
  }
}

And here’s the client setup:

// src/lib/pinecone.ts:

import { Pinecone as PineconeClient } from "@pinecone-database/pinecone";

export const pinecone = new PineconeClient();
export const pineconeIndex = pinecone.Index(process.env.PINECONE_INDEX!);

You can check out the full implementation here: https://github.com/kirandash/ai-web-apps/blob/main/src/app/api/document-query/route.ts

[aulorbe]

Ah okay so fromExistingIndex is a Langchain method?

[kirandash]

Ah okay so fromExistingIndex is a Langchain method?

Yes, that’s correct! fromExistingIndex is indeed a Langchain method. I’ve noticed that the error does not occur when using @pinecone-database/pinecone version ^3.0.3. However, if I update to the latest release (4.0.0 from last week), the error appears, as shown in the previous screenshot.

[aulorbe]

Ahhh okay, thanks for confirming! It is likely that they just haven't updated their integration since our latest major release (4.0.0), but I will confirm shortly!

[kirandash]

Ahhh okay, thanks for confirming! It is likely that they just haven't updated their integration since our latest major release (4.0.0), but I will confirm shortly!

Thanks for looking into this! That makes sense. It’s likely an integration update issue with the latest release. I'll stick with version ^3.0.3 for now, but please keep me posted on any updates or fixes. Appreciate the help! 😊🙏

[aulorbe]

Absolutely, will do! Thanks for bearing with both us and Langchain while we get this sorted :)