FREE Agent OS Engine: Run Agents For $0

**Julian Goldie** (0:00)
Agent Operating Systems are super powerful. I mean, you can basically build and automate any sort of workflow that you want inside a setup like this. So for example, we've even built like a outreach tool with HermesAgent. We've got a voice activated version of Hermes over here. We have a memory galaxy that's ready to go. We can automate videos in like one single click. We can also, for example, automate and create SEO content and deploy it to our websites in one single click. This is all inside one beautiful system that works together to orchestrate and manage our agents. However, one thing that I see people worried about all the time is like, how can you run these agent operating systems for free or get the most out of them? And that's exactly what we're going to talk about today with the free agent operating system engine so you can run a whole operating system of AI agents for free forever using a combination of all the different systems I'm going to show you today.
And the thing I would say here is like a lot of people worried about token usage, they're worried about APIs, they're worried about having a system like this, but then going through their limits on the CLI or their coding plan for the month. And so the whole point of the agent operating system is of course, to let it run and automate autonomously. But if you're worried about doing that, then let me show you some ways around that so you can use free models to help you as much as you can.
So let's get straight into this. And there's basically five systems I'm going to show you today that will help you either completely run this for free or reduce the amount of tokens used massively. So we're going to break down each one of these. As you can see, it's a free local models, free APIs, CLIs, token optimization, and also free memory systems. So you can plug this into your agent operating system. And then you can have these amazing, amazing automations of workflows for running, but you can run them for free. So let's get straight into this.
The first thing that I would say is that you can run local models. And I've tested out quite a few recently. I'll show you what I think is good and what I think is bad, right? And I only have like a, I have a Mac Studio, even then it's not amazing for running local models. But the best one that I've seen so far is the Quen 5, Quen 3.5 27b Coder. That's been a pretty good one that we've used so far. I can create some awesome stuff. Let me show you an example of something that we built here. So this is actually fully built with a local AI agent. We also, for example, created this system, which is not like, I mean, it's not frontier level, right? It's not going to be like Opus 4.8, but if you want to run this for free, and you, you know, for most things, you're not going to be creating like crazy graphics or video games, you're just going to be creating a lot of content and that sort of thing. So if you want to know how to use these, well, you can use the system like I'm showing you right now. And so we have the local engine here, and we can plug in free models into this system, free local models, and then we can build and automate whatever we want. So, for example, that game that just showed you a second ago, that was fully created with this system here. And it's the same with free cloud code. You can run local models and you can run free APIs for free cloud code, which is an open source project that basically takes your free models and then plugs them into the agent harness, which is free cloud code. And so that is method number one, which is using free local models. Now, if you want to indicate where do you get free local models from, you can get them from Huggynface and also Olamma. If you're running local models on a Mac, then you can also use MLX to run them as well. And that seems to be better than Olamma. I've tested them out personally. They've been a lot faster, for example. So that is method number one. If you want a local model to actually be useful, you can check out my local model benchmarks that I've tested pretty much everything with. So you can see we've run 42 different tasks for each of these agents.