@prexo/ai-chat-sdk/dist/services/context/vector-db.js

AI Chat Component with Persistent History

import { DEFAULT_SIMILARITY_THRESHOLD, DEFAULT_TOP_K } from "../../lib/constants.js";
import { nanoid } from "nanoid";
import { BASE_API_ENDPOINT } from "../../lib/utils.js";
function chunkText(text, chunkSize = 500, overlap = 50) {
  const chunks = [];
  for (let i = 0; i < text.length; i += chunkSize - overlap) {
    chunks.push(text.slice(i, i + chunkSize));
  }
  return chunks;
}
class VectorDB {
  index;
  constructor(index) {
    this.index = index;
  }
  async reset(options) {
    await this.index.reset({ namespace: options?.namespace });
  }
  async delete({ ids, namespace }) {
    await this.index.delete(ids, { namespace });
  }
  /**
   * A method that allows you to query the vector database with plain text.
   * It takes care of the text-to-embedding conversion by itself.
   * Additionally, it lets consumers pass various options to tweak the output.
   */
  // eslint-disable-next-line @typescript-eslint/no-unnecessary-type-parameters
  async retrieve({
    question,
    similarityThreshold = DEFAULT_SIMILARITY_THRESHOLD,
    topK = DEFAULT_TOP_K,
    namespace,
    contextFilter,
    queryMode
  }) {
    const index = this.index;
    const result = await index.query(
      {
        ...typeof question === "string" ? { data: question } : { vector: question },
        topK,
        includeData: true,
        includeMetadata: true,
        ...typeof contextFilter === "string" && { filter: contextFilter },
        queryMode
      },
      { namespace }
    );
    const allValuesUndefined = result.every((embedding) => embedding.data === void 0);
    if (allValuesUndefined) {
      console.error("There is no answer for this question in the provided context.");
      return [
        {
          data: "There is no answer for this question in the provided context.",
          id: "error",
          metadata: {}
        }
      ];
    }
    const facts = result.filter((x) => x.score >= similarityThreshold).map((embedding) => ({
      data: embedding.data ?? "",
      id: embedding.id.toString(),
      metadata: embedding.metadata
    }));
    return facts;
  }
  /**
   * A method that allows you to add various data types into a vector database.
   * It supports plain text, embeddings, PDF, HTML, Text file and CSV. Additionally, it handles text-splitting for CSV, PDF and Text file.
   */
  async save(input) {
    const { namespace } = input.options ?? {};
    if (input.type === "text") {
      try {
        const returnId = input.id ?? nanoid();
        await this.index.upsert(
          {
            data: input.data,
            id: returnId,
            metadata: input.options?.metadata
          },
          { namespace }
        );
        return { success: true, ids: [returnId.toString()] };
      } catch (error) {
        return { success: false, error: JSON.stringify(error, Object.getOwnPropertyNames(error)) };
      }
    } else if (input.type === "embedding") {
      try {
        const returnId = input.id ?? nanoid();
        await this.index.upsert(
          {
            vector: input.data,
            data: input.text,
            id: returnId,
            metadata: input.options?.metadata
          },
          { namespace }
        );
        return { success: true, ids: [returnId.toString()] };
      } catch (error) {
        return { success: false, error: JSON.stringify(error, Object.getOwnPropertyNames(error)) };
      }
    } else {
      try {
        let fileSource = void 0;
        if ("fileSource" in input && typeof input.fileSource === "string") {
          fileSource = input.fileSource;
        } else if ("source" in input && typeof input.source === "string") {
          fileSource = input.source;
        }
        if (!fileSource) {
          throw new Error("No valid file source provided for extractText");
        }
        const response = await fetch(`${BASE_API_ENDPOINT}/extractor`, {
          method: "POST",
          headers: {
            "Content-Type": "application/json",
            "Authorization": `Bearer ${process.env.PREXO_API_KEY}`
          },
          body: JSON.stringify({ url: fileSource })
        });
        if (!response.ok) {
          throw new Error(`Failed to extract text: ${response.status} ${response.statusText}`);
        }
        const res = await response.json();
        const transformArgs = "config" in input ? input.config : {};
        const chunkSize = transformArgs && transformArgs.chunkSize || 500;
        const overlap = transformArgs && transformArgs.overlap || 100;
        const chunks = chunkText(res.output.txt, chunkSize, overlap);
        const ids = [];
        for (const chunk of chunks) {
          const id = nanoid();
          await this.index.upsert(
            {
              data: chunk,
              id,
              metadata: input.options?.metadata
            },
            { namespace }
          );
          ids.push(id);
        }
        return { success: true, ids };
      } catch (error) {
        console.error(error);
        return { success: false, error: JSON.stringify(error, Object.getOwnPropertyNames(error)) };
      }
    }
  }
}
export {
  VectorDB
};