chatbot-api

A retrieval-augmented generation (RAG) chatbot library built on LangChain, pgvector, and OpenAI.

What it does

ChatBot wires together an OpenAI-embedded pgvector store, a chat model, and a text splitter into a single class that can:

• Ingest raw text or Documents, splitting them into overlapping chunks and embedding them into Postgres.
• Answer questions grounded only in the ingested knowledge base, with source citations, a confidence rating (high / medium / low), key quotes, and suggested follow-up questions.
• Retrieve with either a plain similarity search or an advanced multi-query strategy, where the LLM rewrites the question into several search queries, retrieves each in parallel, and dedupes the results for better recall.
• Manage conversation history per session, in memory, with automatic trimming to the last 10 turns.

Usage

import { ChatBot } from "./index";

const bot = new ChatBot({
  apiKey: process.env.OPENAI_API_KEY!,
  db: {
    dbHost: "localhost",
    dbPort: "5432",
    dbUsername: "postgres",
    dbPassword: "postgres",
    dbDatabase: "chatbot",
  },
});

await bot.addTexts({ texts: ["Some knowledge base content..."], source: "docs.md" });

const response = await bot.ask("What does the documentation say?");
console.log(response.answer, response.confidence, response.sources);

Setup

yarn install

Requires a Postgres database with the pgvector extension enabled (see init.sql) and an OPENAI_API_KEY.

Project structure

• index.ts — the ChatBot class.
• schemas/ — Zod schemas for structured LLM output (chat responses, multi-query generation).
• init.sql — database bootstrap for the vector store.