How to Build an AI Agent with n8n and Live Google Search Data

n8n is one of the most popular no-code/low-code platforms out there. I've written about n8n once before to cover how to use their generic HTTP node to use our APIs to scrape Google search results. You can find that here:

Today, we'll build an AI agent using the SerpApi tool that ships with all n8n installations. In just a few minutes and clicks, you can have a powerful AI agent that can answer questions requiring live Google search results.

Let's get into it.

Initial Setup

n8n is an open-source platform that you can install and run on your local machine or any server. I'll go for a simple local installation.

You can also opt for n8n's cloud version. This comes with a two-week free trial. After that, you pay $20 per month.

Railway is another popular option and can be run for as little as $5 per month.

Next, you'll need API keys for SerpApi and OpenAI.

You can sign up for SerpApi and get 100 free searches per month here: https://serpapi.com/users/sign_up

Once you're registered, you can grab your SerpApi API key here: https://serpapi.com/manage-api-key

For OpenAI, create an account here: https://platform.openai.com

Unfortunately, OpenAI no longer gives away free credit on signup. As a new user, you'll be able to run a few API calls for free to start, but this won't get you very far. I recommend adding the minimum $5 of credit to your account.

AI Agent Creation

Navigate to the n8n Overview page and click "Create Workflow".

You'll be presented with a blank workflow editor where you can add your first step.

We need to start the workflow with a trigger that kicks off the automation. We'll select "On chat message".

Your workflow should now have a trigger. Hit the "+" button to add the next node.

Here we'll add our agent. Search for "ai agent" and add the "AI Agent" node.

We can leave everything as default except we'll add a basic initial prompt. Click "Add Option" and then select "System Message".

Feel free to put whatever you want here. I'll go with a very generic message.

You are a helpful assistant that helps the user answer whatever questions they have. You will search the Internet for the latest information when required.

We can't quite test this yet because the AI Agent requires a connection to a chat model. Click the "Chat Model" "+" button to add one.

We'll use OpenAI's model.

Here we'll add our OpenAI API key. Click "Create new credential" if you don't have this saved in n8n already.

The form to create credentials is fairly straightforward. You only need to paste in your OpenAI API key. Follow the links n8n provides if you have any doubts on where to get your key.

I'll stick with default model of "gpt-4o-mini".

Now, we can give our agent a test by clicking the "Open chat" button.

Looks good!

Assuming your workflow now looks like the below screenshot, we give our agent some memory to remember past messages. Click the "+" for "Memory".

Let's opt for the simplest "Window Buffer Memory" option since it requires no credentials or complicated setup.

We can leave all the default memory settings.

Now if we test the chat, we can ask follow up questions that rely on context from past interactions.

For example, I've asked "Who invented Bitcoin?" and then I've asked a follow up question that relies on the context of the previous question and answer.

Add SerpApi for Current Google Search Results

So far our agent can only respond with information available to ChatGPT/OpenAI. To get the most current information, we'll add the SerpApi tool.

Hit the "+" button for "Tool".

Search for SerpApi and add it to the workflow.

Similar to OpenAI, the first thing we need to do is add our SerpApi API key as a credential.

Paste in your API key and hit "Save".

You can add a few options to customize your queries to SerpApi, but we'll leave these out for this tutorial.

Let's add one more tool to allow our agent to perform arithmetic with the Calculator tool.

The calculator doesn't require any configuration.

Our workflow now looks like this:

Now if we ask a question that requires current information, we can see in the agent's log that it uses SerpApi to query "current Bitcoin price" and serves us the correct answer!

We can also ask follow up questions requiring the context and a calculation. With this follow up, we can see the Calculator tool is being used, but the SerpApi tool is not since we already have Bitcoin's latest price in memory.

In many cases, we might be happy with text responses, however, we can also ask OpenAI to respond to us in JSON.

Let's update the OpenAI Chat Model node and add an option to modify the "Response Format". We'll select "JSON".

Note the alert above requiring us to include "json" in the prompt.

Let's edit the "System Message" to mention we want the response as JSON.

You are a helpful assistant that helps the user answer whatever questions they have. You will search the Internet for the latest information when required. Give the response as JSON.

Now we can get responses back as JSON.

We'll leave it here for now. I hope you found this useful.

There's lots more you could do such as send the response messages to another system or you could connect more tools to beef up your agent. n8n has tons of pre-built nodes and tools that you can use for whatever you can dream up.

If you build anything cool on top of this, let us know at contact@serpapi.com!

Want to Import This?

If you want to skip all this manual work, you can import the text or JSON output workflows exactly as I've built them in this tutorial. You can then jump in and modify things as you wish. Of course, you'll need to configure your SerpApi and OpenAI credentials in order to run it yourself.

serpapi_ai_agent_text_output

serpapi_ai_agent_text_output.json

3 KB

serpapi_ai_agent_json_output

serpapi_ai_agent_json_output.json

3 KB

To import one of these workflows, create a new n8n workflow and press the "..." button in the upper right. Then select "Import from File."