Enterprise Copilot with External Knowledge

Why Exa for external knowledge?

An enterprise copilot is great at questions about a company's own data — orders, invoices, HR records. But many user questions require knowledge that lives outside the enterprise: general news, people data, company data, and finance. Calling Exa for those questions yields:

Grounded answers: every claim traces back to a retrieved source the user can inspect
Relevance-ranked retrieval: Exa's neural search returns results ranked by semantic relevance, not keyword overlap
Always current: real-time information instead of stale training data
Low latency: a single call returns ranked sources with content highlights — no separate scraping step
Model agnostic: works with any LLM — the copilot's existing model stays in place

Get Started

Install dependencies

bash

npm install exa-js openai

Get your Exa API key from the Exa Dashboard.

You'll also need an API key from your model provider (OpenAI, OpenRouter, etc.).

Initialize clients

javascript

import Exa from "exa-js";
import OpenAI from "openai";

const exa = new Exa(process.env.EXA_API_KEY);
const client = new OpenAI({ apiKey: process.env.OPENAI_API_KEY });

Define the external-knowledge tool

Give the copilot's model a tool it can call when a question falls outside internal data. The tool accepts 1-3 parallel searches:

javascript

const searchTool = {
  type: "function",
  function: {
    name: "web_search",
    description: `Search the external web via Exa for knowledge not available
in internal enterprise data: general news, people data, company data, finance.
Write queries as natural language.`,
    parameters: {
      type: "object",
      properties: {
        searches: {
          type: "array",
          items: {
            type: "object",
            properties: {
              query: { type: "string" },
              numResults: { type: "number", default: 5 },
              category: {
                type: "string",
                enum: ["company", "people", "research_paper"],
              }
            },
            required: ["query"]
          },
          description: "1-3 searches to run in parallel.",
          maxItems: 3,
        },
      },
      required: ["searches"],
    },
  },
};

Exa has dedicated company and people search categories that map directly onto the company-data and people-data question types an enterprise copilot sees.

Run the search and keep the ranking

When the model calls the tool, execute an Exa search. The order of results is the relevance ranking — preserve it so the UI can show ranked sources:

javascript

async function searchExa(query, category, numResults = 5) {
  const start = Date.now();
  const response = await exa.searchAndContents(query, {
    numResults,
    highlights: { maxCharacters: 4000 },
    type: "auto",
    ...(category ? { category } : {}),
  });
  return {
    latencyMs: Date.now() - start,
    results: response.results.map((r, rank) => ({
      rank: rank + 1,
      title: r.title,
      url: r.url,
      publishedDate: r.publishedDate,
      author: r.author,
      text: (r.highlights || []).join("\n"),
    })),
  };
}

highlights returns the most relevant page snippets alongside each result — no separate scraping step needed.

Ground the answer

Feed the ranked results back to the model and stream the final answer. The system prompt instructs the model to answer only from the retrieved sources:

javascript

const followup = await client.chat.completions.create({
  model: "gpt-5.6-luna",
  messages: [
    ...messages,
    {
      role: "user",
      content: `Here are ranked web search results:\n\n${resultsText}\n\nUsing ONLY these results, answer my original question.`,
    },
  ],
  stream: true,
});

Show the sources, not just the answer

Each Exa result includes title, url, publishedDate, and author. Render them as a ranked source panel under the answer, with the retrieval latency.

This gives users the two things an external-knowledge copilot must provide: a grounded answer and the underlying ranked evidence.

Query types this pattern covers

General news & world knowledge: “What are this week's biggest developments in enterprise AI?”
People data: “Who leads a major supply-chain organization and what have they said recently?”
Company data: “Give me an overview of a leading cloud software company and its recent announcements.”
Finance: “How have global semiconductor markets performed this year?”

Conclusion

The copilot's model decides when a question needs external knowledge, Exa retrieves relevance-ranked sources with content in a single call, and the UI returns a grounded answer with the ranked evidence and latency attached.

Get started with Exa for free at dashboard.exa.ai.