How to build an AI agent with n8n (Step-by-step guide)

Building an AI agent with n8n lets you automate workflows, connect Large Language Models (LLMs) like OpenAI or Google Gemini, and create intelligent systems that respond to user input. To do this, you’ll set up a workflow with a trigger, add an AI Agent node to serve as the agent’s “brain,” and connect it to an LLM using an API key. You can also extend its capabilities by adding memory for context retention and tools for tasks like searching the web or sending emails. Finally, the agent’s behavior is defined through a system prompt, and you can test it directly in n8n’s chat interface.

In this step-by-step guide, we’ll show you how to:

1. Install n8n on your VPS
2. Set up a workflow trigger
3. Add and configure the AI Agent node with an LLM
4. Define system prompts, memory, and tools
5. Test, debug, and refine your AI agent

By the end, you’ll have a fully functional AI agent that you can expand and adapt for your own use cases.

Prerequisites

Before getting started, make sure you have the following:

• Hostinger VPS plan based on resource usage
  • Minimum: 1 vCPU, 1GB RAM (KVM1)
  • Recommended: 2 vCPU, 2GB RAM (KVM2)
  • If you don’t yet have one, Hostinger offers VPS for n8n, all of which include a single-click n8n template
• Basic understanding of how n8n works
• A working email address for n8n setup and free license activation

With the prerequisites in place, you’re ready to install n8n on your VPS and set up the foundation for your AI agent.

1. Install n8n

n8n installation sets up the environment needed for building your AI agent on a VPS. Select the n8n template from the Operating System tab in the VPS dashboard. This ensures that all dependencies are properly configured, so you can start building workflows right away.

Once the installation is complete, access n8n from your VPS dashboard on the Overview page by clicking the Manage App button, which will bring you to the login page of your n8n instance.

During setup, you’ll need to enter your user details by providing your email address, full name, and password that you’ll use later to access your n8n account.

After the successful sign-up, you’ll be greeted with two pop-ups: a brief questionnaire and an offer to unlock premium features.

To unlock the free premium features, which include workflow history, advanced debugging, and more, just enter your email address. You’ll be sent a license key.

You can then enter the license key under the Settings → Usage and plan tab.

Another important feature to enable under the Settings → Personal tab is 2-factor authentication (2FA). This adds an extra layer of protection, keeping your n8n account more secure.

Once n8n is installed and configured, the next step is to create a workflow trigger that activates your agent.

2. Set up your workflow trigger

Every automation needs a starting point. In n8n, that starting point is a trigger – an event that tells your workflow to start. Whether it’s running on a schedule or receiving data from an external source – such as a new message – the trigger is what sets everything in motion.

To start, go to the Workflows tab, and click Create Workflow.

The first thing you need in your workflow is a trigger node. To add a trigger, just click the + button and choose the one that fits your needs.

For AI-driven workflows, the most commonly used triggers are:

• App event trigger – This is useful when your AI agent needs to react to messages or notifications from platforms like Slack, Telegram, or email. For example, if a user sends a message in a Slack channel, the AI agent can process the text and respond accordingly.
• Webhook trigger – Ideal for real-time automation when integrating external services. Here n8n generates a unique webhook URL that can receive data from an API, a chatbot, or any other system. Once the webhook receives a request, the workflow runs instantly.

In this tutorial, we’ll use the Chat message trigger node type because it is a simple way to test your first AI agent automation.

Once your trigger is in place, you can connect your AI model and define what actions it should take.

3. Add the AI agent service

Next, it’s time to add an AI Agent node to your workflow. The AI Agent node allows your workflow to communicate with an external AI service to process text, generate responses, or analyze data.

If you’re already familiar with API integration in workflows, you might consider integrating the AI service using an HTTP Request node. This gives you full control by manually setting the endpoints, headers, and payloads.

However, the AI Agent node offers a more straightforward and user-friendly alternative. It takes care of handling the API calls for you, allowing easy connection to services like OpenAI or Google Gemini with minimal configuration. We will use this type of node for this tutorial.

To add the AI Agent node click the + icon next to your trigger node, search for AI Agent, and add the node to your workflow.

Once added, you’ll need to configure its core settings at the bottom of the node: Chat Model, Memory, and Tool.

Chat Model is required for your AI Agent node to work. In this tutorial, we’ll use Google Gemini because it offers a free tier, but you can use any of your preferred chat models.

Press the + button at the bottom of the node. Then, set up the Google Gemini (PaLM) credentials by inserting the API key that you can generate in Google AI Studio. During this stage, you can select your preferred chat model, such as Gemini 1.5 Pro.

Next, attach a Memory sub-node if you want the AI to retain context across multiple interactions. This works as a database for storing conversation history, making it useful for chatbots and ongoing conversations.

Without it, the AI would treat each request separately, losing the previous context – so the following interaction would fail:

User: Who won the last World Cup?
AI: Argentina won the 2022 FIFA World Cup.
User: What country hosted it?
AI: I’m not sure what you’re referring to. Can you clarify?

The Simple Memory option is the most straightforward because it uses internal n8n memory and does not require additional setup or credentials.

Also, the AI Agent node allows you to add a Tool sub-node – this enables the AI to interact with external applications.

For this tutorial, add the SerpAPI tool sub-node, which we will use to extend our AI capabilities by allowing it to search the web. Once added, you’ll need to register for a free account and retrieve your API key to set up node credentials in n8n.

4. Build and customize the AI agent workflow

Now that the AI agent node is set up with a chat model, memory, and search capability, the next step is to refine the workflow by adding decision-making logic and optimization techniques.

You can start by expanding how the AI processes different types of messages. To do this, add a Switch node after the trigger. This node routes the input data to specific nodes based on keywords or conditions. For example:

• If a message contains the word search, the workflow prioritizes the SerpAPI tool for retrieving real-time web results.
• If the conversation is part of an ongoing chat, ensure that Simple Memory is properly handling context so the AI can remember previous exchanges.
• If no special conditions are met, the message is sent to Google Gemini for standard processing.

This can be achieved using regular expressions (regex):

Use SerpAPI and Use Memory nodes should connect to the AI Agent through the Edit Fields nodes, while Regular query connects directly to the AI Agent.

The Edit fields nodes will add the following directions to the chat input field:

• Use SerpAPI – The user wants real-time information. Use SerpAPI to find the latest answer. The user asks: {{  $json.chatInput }}

• Use Memory – You are in an ongoing conversation. Use memory to recall past interactions. The user says: {{  $json.chatInput }}

When it comes to handling workflow errors, the default setting of each node is to halt execution if an error is encountered. This approach ensures anomalies like malformed data are not passed along to subsequent steps.

If you want to be notified when an error occurs in your workflow, you can create an additional Error Workflow, which is automatically triggered when your regular workflow runs into an error.

To create an Error Workflow, simply click the + icon at the top left corner of the page, select Workflow, and name it Error Workflow.

In the new workflow, add the Error Trigger node. You can connect it to any application node you want to use for notifications, for example, Gmail.

Back in the AI Agent workflow you can open Workflow Settings and set the Error Workflow to the one you have just created.

Now, if your workflow runs into an error you’ll get an email notification, ensuring you can promptly address issues in your automation.

5. Test and debug the workflow

At this point, a fully functional AI agent is now integrated into your n8n workflow. You can test it by clicking the Open Chat button at the bottom of the screen and sending a sample message.

When testing, send different types of messages to make sure the AI responds appropriately based on the input. Notice how different workflow routes are triggered when specific keywords are detected.

For example, try asking a generic question and wait for the response, then follow up by asking your AI Agent if it can remember what it told you previously. You can see the triggered workflow path highlighted in green.

If you receive a notification that your workflow ran into an error, you can review and debug in the Executions tab.

To find out where your execution failed, look for the node marked in red with a warning icon. Double-click the node to see the full error message, which often includes an explanation of exactly why the error occurred and a hint on how to fix the issue.

In this example, the error is caused by a common issue: a mismatch between the expected and actual field types. The sessionId field is received as a string, but the Edit Fields node expects a boolean. This can be fixed easily by changing the expected field type to string.

What can your n8n AI agent do next?

Now that your n8n AI agent is up and running, the real opportunity lies in expanding what it can achieve. Beyond simple message processing, your agent can manage emails, trigger web searches, process data from APIs, or even coordinate tasks across multiple apps. Each of these capabilities helps transform your agent from a basic chatbot into a versatile automation assistant.

To see the full range of possibilities, check out our guide on what you can automate with n8n. It explores practical use cases that will help you unlock more value from your AI agent and scale your automations with confidence.

All of the tutorial content on this website is subject to Hostinger's rigorous editorial standards and values.