upsertFromYouTube

Inserts or updates vectors from a YouTube video.

upsertFromYouTube(
    videoId: string,
    options?: {
      language?: string
      addVideoInfo?: boolean
      metadata?: Record<string, any>
      textSplitter?: SplitterParams
    }
): Promise<string[]>

Reference

import { myVectorStore } from "#elements";
export default async function () {
  const count1 = await myVectorStore.upsertFromYouTube(
    "https://www.youtube.com/watch?v=bZQun8Y4L2A",
    { language: "en", addVideoInfo: false }
  );
  console.log(`${count1} vectors upserted`);
  const count2 = await myVectorStore.upsertFromYouTube("bZQun8Y4L2A", {
    language: "zh-Hans",
    addVideoInfo: false,
  });
  console.log(`${count2} vectors upserted`);
}

Parameters

• videoId: The ID of the YouTube video to extract content from.
• options: Optional configuration parameters, including:
  • language: (optional) The language of the subtitles to extract, defaults to 'en'.
  • addVideoInfo: (optional) Whether to add video information (title, description, etc.) to the metadata, defaults to false.
  • metadata: (optional) The metadata to associate with the vectors.
  • textSplitter: (optional) The text splitter employed to divide the content into multiple vectors. In the absence of a provided splitter, the token splitter is used by default.

Returns

Promise of an array of IDs of the upserted vectors.

Caveats

• This method will insert a new vector if the videoId does not exist, or update the existing vector if the videoId exists.
• When addVideoInfo is set to true, there may be instances where the video metadata is not available or accessible.
• You can query all the results by filtering the metadata field source-by-babel to videoId.

Examples

In the HTTP Element, load YouTube video transcript data:

import * as Koa from "koa";
import { myVectorStore } from "#elements";

export default async function (
  request: Koa.Request,
  response: Koa.Response,
  ctx: Koa.Context
) {
  await myVectorStore.upsertFromYouTube("dVtrJYSdVo0");
  const result = await myVectorStore.search("How to cook Steak", {
    topK: 3,
  });
  return result[0].metadata!["transcript"];
}