Add live web search to a Node chatbot with Tavily

What you get

A Node function that searches the live web with Tavily and returns titles, URLs, snippets and an LLM-generated answer to ground your chatbot's replies.

Prerequisites

Node.js 18+
A Tavily API key from https://app.tavily.com (1,000 free API credits every month, no card required)

Steps

Install the SDK:

Shell

npm i @tavily/core

Set your key:

Shell

export TAVILY_API_KEY="tvly-YOUR_API_KEY"

Create search.js:

JavaScript

const { tavily } = require("@tavily/core");

const tvly = tavily({ apiKey: process.env.TAVILY_API_KEY });

async function webSearch(query) {
  const response = await tvly.search(query, {
    maxResults: 5,
    includeAnswer: true,
  });
  return {
    answer: response.answer,
    sources: response.results.map((r) => ({ title: r.title, url: r.url, content: r.content })),
  };
}

webSearch("Who is Leo Messi?").then((r) => console.log(r));

In your chatbot, call webSearch(userMessage) (or expose it as a tool) and paste sources into the system prompt so the model can cite them.

Test it

Shell

node search.js

You should see an answer string and up to 5 sources with title, url and content.