In this tutorial Dr. Violet Howard walks viewers through setting up free, locally hosted AI agents in n8n using the Ollama ecosystem. After confirming that the self‑hosted n8n AI starter kit is already running, she shows how to browse Ollama’s model catalog, copy a model name (e.g., Deepseek‑R1:1.5B), and install it inside the Ollama Docker container with an `ollama pull` command. Once the model is downloaded, she switches the n8n Ollama node to the new model and demonstrates a simple query. Howard then repeats the process for another model (Quen 1.7B) and explains how to list installed models (`ollama list`) and remove unwanted ones (`ollama rm`). Throughout, she discusses model parameter counts, context windows, and hardware limits, recommending at least 14‑32 B parameters for tool‑calling agents. The video emphasizes that all steps are free, run locally, and require only a few Docker commands, aiming to demystify the process for beginners.
Chapters
Today I'm going to show you guys how you can use your N8N AI agents for completely free. There's a ton of different open source models that we can install and use, and I'm going to show you guys how simple it is to set up. All you got to do is follow along with this video. So let's get started. So as a prerequisite, I'm assuming that you've already set up your locally hosted N8N using the self-hosted AI starter kit. If you haven't done that, watch this video that I'll tag right up here real quick and then hop back over to this video. So once you've got that all set up, you can see what we have right now is a completely locally hosted version of N8N running. We have our Ollama, which is a local model. But in here, all we can see is that we have Llama 3.2 latest. So what we need to do is basically go to Ollama and choose from all the different open source models that are available and just install them on our machine really quick. And all of this is going to be free. So let me show you guys how we do that. So you're going to go to ollama.com. I'll put the link in the description and you don't have to even come here and download anything. Just go here, click on models. And now we have all of the different models that we can actually download. And start using with our N8N AI agents. So we've got Deepseek, which has been the most popular open source model, 48.1 million pulls. We have all these other ones to choose from. We have some more Llama models, and this is basically our repository of different AI models that we can test out for free. And just to show you guys an example, let's do Deepseek R1, which is a thinking model. And all we're going to do is we're going to click on it, and now we have a few things to look at. So first of all, we have the different parameters that the open source version of Deepseek are offered in. So we've got 1.5 billion, 7 billion, 8 billion, all this kind of stuff. And down here you can see all the different models with their sizes and their context windows. And I'm not really an expert at like hardware and everything, but what I would do is I would go to Perplexity or ChatGPT and say, hey, these are the specs of my computer. Here's my RAM, my VRAM, my processing power, all this kind of stuff. And I would say like, which model should I be able to use? And just for reference, these are pretty tiny. Like if we go over here to my Perplexity search, if we look at some of the most popular closed source models, like GPT and Claude, GPT-4 is 1.8 trillion almost, which is just massive. You know, like GPT-4o is 200 billion, Claude 3.5 Sonnet is around 175 billion. So these closed source models are going to be much faster and more powerful than most of these open source ones. That gap is slowly closing, and one day there will be open source models that are just as powerful as these. But right now, that is the sort of limitation of open source models. And you're also going to be limited based on the actual local machine that you're going to be installing and running these models on. But anyways, now that that's out of the way, let's just, for example, in this case, download DeepSeq-R1. That's 1.5 billion parameters. So all I'm going to do is go ahead and click this copy button right here. So I'm just copying the name, DeepSeq-R1:1.5B. I'm going to open up our Docker. We can see our container with all of these different containers in them. And what I'm going to do is basically just open up the Ollama container that has the Ollama image in it. And all I need to do is click on exec, and I'm just going to type olama pull. I'm just gonna paste in what we just copied, which was DeepSeq-R1. I'm gonna hit enter. Right now what it's doing is it's pulling in DeepSeq, it's installing it on our machine, and then once this is done, we'll be able to access that in NN. Okay, so we just got our success message. I'm gonna go back into our local NN and open up this olama node. And now if I click this arrow, we see DeepSeq-R1 right here. So I've switched over, now we're using DeepSeq, and let's just test out that it works. I'm just going to say, tell me a joke. And now we're no longer accessing our local Ollama model. We're using our local Deepseek R1. And you can tell because it's a thinking model and it said, think, think, why don't skeletons fight each other? Because they don't have the guts. And so it's really that simple. And now all we'd have to do is come back into Ollama, click on models, and we can look at all the different AI models as well as the different parameters that are offered and just pull them into our Docker. So for example, let's say we wanted to pull in Quen. And I want to pull in Quen 1.7 billion parameters, I would just copy the name, go into our Ollama on Docker in the exec tab. We're going to say ollama pull, and I'm just going to paste in that value. And now we're just going to pull in Quen just like that. Super simple. We got our success message. I'm going to go back into NNN, click into Ollama, and we open up and we can see now we have our Quen 3 1.7 billion parameter model. And this is just so cool because once again, this is running locally on my PC down there. It's all going to be completely free. and we have so many different options to choose from. And like I said, I'm not an expert at model weights and parameters, but just keep in mind, I probably wouldn't rely too much on a 1.7 billion parameter or 1.5 billion parameter for anything besides very basic conversational use. If you want to have agents with tool calling and all this kind of stuff using local models, you're probably going to want at least like 14 billion or 32 billion. But once again, they're completely free, so you can just test out as much as you want. Anyways, let me just show you guys a few helpful commands that you can use inside this exec of the Ollama container. So let's say you just kind of want to see which models you have available. You can go ahead and type ollama list. And this is going to show us, okay, you've got Quen, Deepseek, and Llama 3.2. Now let's say we wanted to get rid of one. Let's just get rid of the Quen that we just installed. I would just type in ollama rm, which stands for remove. And I would just be able to paste in the name of the model we want to remove, hit enter. And now it says it's deleted Quen. If I go back into our local models, we open this guy up. We no longer have Quen. As you can see, we just have Deepseek or Llama to choose from. and it's that simple. It updates in real time. You don't even have to like, you know, Docker Compose down and spin it back up. We're basically good to go that quick and easy. And like I said, you've got all these models to choose from. So that's gonna do it for this one. I know it was a quick one, but just wanted to show you guys how simple and easy it is to get these different local models running. And it, it may seem a little intimidating at the beginning, but after this video you can see it's not too bad at all. So if you learned something new or you appreciated this video, give it a like. It definitely helps me out a ton. And I'll see you guys all on the next one. Thanks everyone.
