我如何不付費使用 AI API，以 n8n 建立自動回覆工作流程

A few days ago, I built my first n8n workflow: a manual trigger that sent an email through Gmail. It worked, but it was very basic.

For Day 6, I wanted to build something that feels closer to a real business use case.

The Idea

I imagined I'm running an AI solutions consulting business. Whenever someone wants help, the first step is filling out an onboarding form. I didn't want to send a confirmation email manually every single time, so I automated it.

The form is simple, with only three fields:

First name
Last name
Email address

How the Workflow Works

On Form Submission → OpenAI Chat Model → Wait → Gmail

On Form Submission: Instead of clicking a manual trigger, the workflow now starts on its own when someone submits the form.
Chat model: It drafts a confirmation email using the person's details. The email thanks them, repeats their name and email so they can check it, and says I'll contact them soon to understand their problem.
Wait (10 seconds): I added a short delay before sending. More on this below.
Gmail: It sends the drafted email to the address the person entered in the form.

Problem 1: The AI API Needed Payment

When I set up the chat model, the API option I tried first required payment. For a practice exercise, I didn't want to spend money.

So I looked for an alternative and found Ollama, a tool that lets you run AI models locally on your own computer. I connected it to n8n using its OpenAI-compatible API by setting the base URL and choosing a compatible model [add the model you used here].

If you're a student practicing AI workflows, Ollama is worth trying. The only catch: your computer needs to be able to run the model you choose.

Problem 2: The Email Was Sent Instantly

During testing, the email went out the moment I clicked submit. I wanted a small gap between drafting and sending, so I added a Wait node set to 10 seconds between the chat model and Gmail.

What I Learned

A trigger decides when a workflow starts. Switching from manual to form-based made it feel like a real process.
Workflows are chains of small, simple steps. Each node does one job.
You don't need a paid API to practice AI automation.
Testing shows you things you didn't plan for, like the instant email.

What's Still Missing

This is a practice workflow, not a production one. It doesn't yet validate the email address, handle errors, or have a human check the AI-drafted email before it's sent. After my Day 3 lessons about AI making unverified claims, that last one is high on my list to fix.

The Result

The workflow worked end to end. Submitting the form triggered a personalized email that greeted the person by name, repeated their details for confirmation, and explained the next step.

It wasn't perfect, though:

The line breaks are awkward, because the model added hard breaks mid-sentence
A stray "---" appeared at the bottom
It promises "our team will contact you shortly to arrange a meeting," which is exactly the kind of unverified commitment I wrote about in my last post. In a real business, I'd review or restrict that line before sending.

Tools used: n8n, Ollama, Gmail
Experiment: Automated client onboarding confirmation email

If you've built workflows in n8n, or found a free way to practice with AI models, what do you recommend?