AI Workflows

Create an AI Support Agent

Build an autonomous support agent using tools and function calling.

Build an AI support agent capable of answering customer questions, retrieving information, and performing actions using tools and function calling. This guide demonstrates how to create an intelligent assistant for customer support workflows.

Overview

Unlike a traditional chatbot, an AI support agent can reason about user requests, retrieve relevant information, and use external tools to complete tasks such as checking order status, creating support tickets, or searching a knowledge base.

In this guide, you'll create a simple support agent using NovaAI Agents.

Estimated time: 15–20 minutes

Prerequisites

Before you begin, make sure you have:

  • A NovaAI project

  • API authentication configured

  • Basic knowledge of Agents and Function Calling

  • Access to the NovaAI Agents API

Start with one or two tools before building more complex support workflows.

Start with one or two tools before building more complex support workflows.

Agent Workflow

Rendering diagram…
1

Create the Agent

Initialize a support agent with clear instructions.

const agent = await client.agents.create({
model: "nova-agent-1",
instructions: "You are a friendly customer support assistant that provides accurate and concise answers."
});
1

Create the Agent

Initialize a support agent with clear instructions.

const agent = await client.agents.create({
model: "nova-agent-1",
instructions: "You are a friendly customer support assistant that provides accurate and concise answers."
});
2

Register the tools your agent is allowed to use.

2

Register the tools your agent is allowed to use.

Tool

Purpose

Knowledge Base

Answer product questions.

Order API

Check order status.

Ticket System

Create support tickets.

FAQ Search

Retrieve common answers.

Only expose the tools your application actually needs.

3

Handle Customer Requests

When a customer sends a message, forward it to the agent.

const response = await client.agents.run({
agent: agent.id,
message: "Where is my order?"
});
 
console.log(response.output);

If additional information is required, the agent can call one of the connected tools before responding.

3

Handle Customer Requests

When a customer sends a message, forward it to the agent.

const response = await client.agents.run({
agent: agent.id,
message: "Where is my order?"
});
 
console.log(response.output);

If additional information is required, the agent can call one of the connected tools before responding.

Example Conversation

Customer:

Where is my order?

Agent:

Let me check that for you.

✓ Order #20481

Status: Shipped

Estimated Delivery: Tomorrow

Recommended Agent Configuration

Setting

Recommendation

Model

nova-agent-1

Temperature

0.3

Memory

Enabled

Streaming

Enabled

Function Calling

Enabled

Best Practices

  • Give the agent clear responsibilities.

  • Connect only trusted tools.

  • Keep responses concise and professional.

  • Escalate complex issues to human agents.

  • Monitor conversations to improve performance.

Design your agent to acknowledge the customer's request before taking action. Even a brief response like "Let me check that for you..." makes interactions feel faster and more natural.

Design your agent to acknowledge the customer's request before taking action. Even a brief response like "Let me check that for you..." makes interactions feel faster and more natural.

Completed

Congratulations!

You've built an AI support agent capable of answering questions and interacting with external tools to assist customers.

Was this helpful?

Was this helpful?

Create a free website with Framer, the website builder loved by startups, designers and agencies.