How To Run 5 AI Agents for FREE (OpenClaw Full Setup) transcript

Last month, I went through my credit card statement, and I found something that genuinely made me feel stupid. I was paying [music] $1,003 a month for 10 different software tools. Three years of auto pay subscriptions just quietly [music] drained $12,000 a year out of my business. Then, within 1 week of setting up my AI agents, I canceled almost all of it. I now pay $140 a month while [music] my agents make me over $31,000 every month.

And in this video, I'm going to show you exactly >> [music] >> how to set this up for free, for $30 a month, or for what I personally run. Three tiers, three setups, you pick whatever works for you. So, without further ado, let's get into [music] it. Okay, some quick context. If this is your first time watching me, I'm Sharbell. I run Unfungible, a marketing agency, and I've been building AI agents for about a year now.

I currently run five agents on a single Mac mini in my office. They work 24/7. They report back to me every morning. They've replaced a full-time team's worth of work. My five agents are We have Max first, who's my chief of staff. He manages all the other agents and reports everything back to me. He's the only one I talk to. He's the one I send Telegram messages back and forth to. >> [snorts] >> Then, we have Sage, who's my ex content specialist.

He grows my Twitter presence, generates tweet ideas every morning based on what's trending. You can see him here generating a bunch of tweet ideas for me to go through. And then, we have Nova, my YouTube strategist. She grew my channel from 600 to 2,500 subscribers in 4 weeks. Something I haven't been able to do in a year of posting on YouTube. Over here, you can see my analytics. Guess when Nova started helping me out.

Just take a wild guess. Then we have this little guy, Knox. He's my trading bot monitor. He tracks my PolyMarket and HyperLiquid bots and makes me about $400 to $500 every single month. You can see him operating here, making his own trades, and yeah, he currently has a 96% win rate, and he's made $2 this far today. In the 15-minute market bot, he's made $5 so far. In the 5-minute market bot, you understand. And I've just woken up.

This is literally just me waking up, and he's made me money. And lastly, but not least, we have this little guy, Pixel. He's my product and web apps developer. He helped me generate over $30,000 last month managing my businesses' web apps. I have a full video on my AI agents, if you're interested, but today we're focused on one thing. How to actually set this up at whatever budget you have with three different tiers depending on your budget.

Let's go. Okay. So, for tier one, this is completely free. This is for you if you have a spare device lying around, like an old laptop, an old desktop. It doesn't need to be fast. It just needs to be able to stay on all the time. Step one, you will want to install Open Claw. Go to openclaw.ai and follow the installation. Here, you have a quick start. It takes about 10 minutes to install, and the link is in the description.

Step two, you'll want to install Ollama. Ollama lets you run your AI models locally, directly on your machine with no API cost, no subscription, no cloud. The AI runs on your device. You literally go to Ollama, select your operating system, and click on download. That's it. And then through Ollama, download Qwen 3.5. This is the model your agent will use to think. You'll want to go here and find the Qwen that suits you and suits your own operating device.

The good thing about Qwen, it's completely free, it's capable, and for basic tasks, it handles pretty much everything that you need. Or, you can literally just ask it, which Qwen is best suited for my device? That's what I did, and it just started downloading it. But once you successfully do that, you move on to step four, which is connecting Ollama to your Open Claw. In Open Claw settings, if this is your first time installing it, pick Ollama as your provider.

If you already have Open Claw installed, then switch the model provider to your local Ollama instance. That's it. You now have a fully operational AI agent running for $0 a month. The one trade-off though is that your device needs to stay on 24/7. That's the cost of completely it being free. If you have a device that's been sitting in a drawer, you can just plug it in and leave it running. And that's it. You're done.

The limitation for you to keep in mind though is that local models like Qwen 3.5 are good, they're not great. For simple tasks like scheduling, summarizing, research, drafts, they totally do the job, but for deep reasoning, like complex coding or even high-quality writing, you will want either tier two or three. Okay, let's move into tier two, the cost-effective setup, which is about $10 to $30 a month. This is the sweet spot for most people.

If you have a device laying around, you can use it here, too. Or, if you don't have a spare device, then I'll point you to what you can use in just a bit. For this specific setup, you will want to use Open Router, which gives you access to the best AI models at API pricing. Meaning, you don't pay a monthly subscription, you pay per use, and the costs are tiny. The model I point you to here is Kimmy K2, one of the best models for the price right now.

It's fast, intelligent, and cheap. It's great for the kinds of tasks you'd be running on agents. For the hosting side, that's if you don't have a device that you can use, instead of a device at home, you can get a VPS subscription. Some people call this a cloud server. The costs for this are about $5 a month, which is a subscription that you pay every month, and you have providers like Hetzner, Digital Ocean, and Vultr.

The one I personally use is Digital Ocean, but you don't have to use that. That's just what I'm used to, what I've tried, and what I've gotten used to over the years. Any of them realistically work. I mean, you install Open Claw on the VPS, you connect it to Open Router, drop in your API key, and that's it. You're running it. The total cost for this is somewhere between $10 to $30 a month, depending on your usage, and depending on which AI models you end up choosing to run your agents on.

The upgrade here over tier one is that you're not dependent on a device at home staying on. The VPS is in the cloud. It runs 24/7 without you touching it. And the model quality, since we're using this on Kimmy's AI instead of Quen, is significantly, much better. Trust me. And this is where we get to the final tier, tier three, what I personally run. This is for people who want the best performance. It's what I run. It's what I'd recommend if you're using agents seriously, and you want real results.

For hardware, I use a Mac mini that I paid $600 for. But, you don't need a Mac mini. I mean, any dedicated device works just fine. The point is dedicated hardware that runs 24/7 and isn't your main machine. And in terms of AI models, and this is the part most people get wrong, different tasks need different models. Treating every single task the same is how you get mediocre output. So, here's my exact stack. For conversation and back and forth with my agents, I use Claude Sonnet.

It's fast, it handles nuance as well, and it's great for operational decisions. For coding and builds, I use Claude Opus or sometimes GPT 5.4. When I'm shipping code, I want the best model on it. So, don't cut corners here. You will notice the difference if you use most of the other AI models out there. For image generation, I use Gemini's Nano banana. That's the only place I use Gemini. If you've used Gemini before for anything other than image generation, I I salute you.

But, it is the best quality to cost for visuals so far. And then for X research and tweet analysis, I use Grok's API. It has native access to live X data, which no other model comes even close to for the specific use case. The total cost for me ends up at roughly $120 to $200 a month in API costs and subscriptions. Now, let me tell you why I call this basically free. My trading agent made me $400 last month just from trading.

My X agent helped me earn $800 in 2 weeks from creating X content, so from X creator revenue. >> [snorts] >> My YouTube agent grew my YouTube channel from 600 to 2,500 subscribers in 4 weeks, and I'm on track for monetization. Finally, this pen literally pays for itself multiple times over. And here's a big bonus for you if you're still watching this. Here's the thing nobody tells you when you first start building agents.

You can cancel your software subscriptions. I was paying $1,000 a month across 10 tools. My agents replaced every single one of these subscriptions. My YouTube agent replaced Vista Social. She doesn't just schedule, she actually writes the content now. My X agent replaced Brand24 with a $12 a month Grok API called that gives me much better intel than ever before. My trading agent replaced TradingView. He reads the logs, but not only that, he tells me why something's off, not just that it is.

Agent memory and my mission control replaced Notion. My X agent replaced ActiveCampaign. My main agent replaced Inbox app and Typeform. You get where I'm going with this. In total, I've saved $863 a month or $10,356 every single year. The agents paid for themselves in week one. So, that's the full breakdown. Three tiers. The free one, which is you using OpenClaw, Ollama, and Qwen 3.5 on any spare device. The 10 to 30 dollars a month one, which is you using Open Router, Kimiko 2, and a 5 dollar VPS subscription.

And the third one, the 100 to 300 dollars a month one, which is you using dedicated hardware and the best model you can for every task, which is what I basically run. If you want to see exactly what each of my five agents does day-to-day, that video is right on my channel. Or if you want to grow your business's content pipeline and in turn grow your revenue, we also help set up these exact workflows for you and your business.

So, make sure to book a call with us through our website. The link is also in the description. Don't forget to subscribe if you want more of this. I post every single week and I'll see you in the next one.