Setup
Copy
Ask AI
npm i openai @nangohq/node
This example uses the Node SDK, a lightweight wrapper over Nango’s REST API. The same flow works in any language by using the API reference.
- A Nango account with a Hubspot integration configured and the
whoamiaction enabled. - An OpenAI account.
Usage
Example of a basic tool that calls an external API (Hubspot):Copy
Ask AI
import OpenAI from "openai";
import { Nango } from "@nangohq/node";
export async function runOpenAIAgent() {
const client = new OpenAI({ apiKey: process.env.OPENAI_API_KEY! });
const nango = new Nango({ secretKey: process.env.NANGO_SECRET_KEY! });
const userId = "user1", integrationId = "hubspot";
// Step 1: Ensure the user is authorized
console.log("🔒 Checking API authorization...")
let connectionId = (await nango.listConnections({ integrationId, userId })).connections[0]?.connection_id;
// Step 2: If the user is not authorized, redirect to the auth flow.
if (!connectionId) {
const session = await nango.createConnectSession({
allowed_integrations: [integrationId],
end_user: { id: userId },
});
console.log(`Please authorize the app by visiting this URL: ${session.data.connect_link}`);
const connection = await nango.waitForConnection(integrationId, userId);
connectionId = connection!.connection_id;
}
// Step 3: Send a prompt to your model.
console.log("🤖 Agent running...")
const response = await client.responses.create({
model: "gpt-5",
input: "Using the who_am_i tool, provide the current user info.",
tools: [
{
type: "function",
name: "who_am_i",
description: "Get the current user info.",
parameters: { type: "object", properties: {}, additionalProperties: false },
strict: true,
},
],
});
const functionCall = response.output.find((item) => item.type === "function_call") as any;
if (functionCall?.name === "who_am_i") {
// Step 4: Execute the requested tool with Nango.
const userInfo = await nango.triggerAction(integrationId, connectionId, "whoami"); // Call tool via Nango.
console.log("✅ Agent retrieved your Hubspot user info:", userInfo);
} else {
console.log(response.output_text);
}
}
You can let agents or models introspect available tools and their specifications by calling this endpoint. Your model can then automatically decide which tool to call.