Claude + Ruflo Agent Swarms is Insane (FREE!)

Claude just got a free upgrade that makes it 100 times more powerful. In this video, I'm going to show you exactly what it looks like using something called Rufflo and Agent Swarm. So when you combine it with Claude, you don't just get one AI agent working for you. You get a hundred, a hundred specialized AI workers all running at the same time, all coordinating together, all getting smarter the longer you use them.

Think about what that actually means for your business. Tasks that used to take you hours are now done whilst you sleep. Content, research, outreach, all running in parallel as AI agents without you touching a thing. And the best part is it is free to set up and use with Claude Code.

I'm going to show you how to set this up step by step, what it actually does in real time. And the one thing that makes Rufflo different from every other AI tool you've ever tried, which is what I'll get into in just a minute. So today we're going to be testing out Rufflo, which is a multi-agent AI orchestration for Claude Code, where you can orchestrate specialized AI agents. You can actually orchestrate like a hundred of them.

I'm going to be testing this out with Claude Code directly and seeing how it performs, right? So you might be wondering, okay, what's the difference between the two, between just using Claude Code and using Rufflo? So with Rufflo, you get a few agents, you can manage a few slash commands, et cetera. With Rufflo, you get the full Rufflo loops and 98 agents, 60 commands, 30 skills and MCP server hooks and everything else, right?

You also get files inside your workspace, MCP server set up and it installs hooks as well for you. And it's an open source project. So here's what you get with Rufflo, right? Number one, 100 agents specialized for coding, testing, et cetera.

You get a comms layer, you get swarm coordination, which means you can manage your agents swarms together. It's self-learning like news research. So it can actually self-improve as well. There's a vector memory, background workers, and also a bunch of other stuff too, like a web UI beta.

So we're going to test out now with Claude and see how this performs. So basically it just makes Claude more autonomous and powerful because you have AI agents that can work in orchestration, right? That can work together. So we're gonna get started with this now, and I'm just going to go into terminal here and set up Rufflo with Claude Code.

And it's just installing here, as you can see. So let's talk about this and what it means, et cetera. So basically you can put like a hundred AI workers to work, right? Using Rufflo, which is super powerful and pretty crazy when you think about it.

So Rufflo is basically a platform that lets you control a whole army of AI agents that coordinate together, right? And it runs on top of Claude Code. So it makes it way more powerful as a tool, right? Now you might think, okay, why should you care?

It's great for saving time. And basically it automates tasks that teams could do. It learns from every task to get smarter over time. There's multiple agents working in parallel, not one by one.

It remembers things across sessions as well, and it connects to the tools that you already use. So how does it actually work? You want to think of it like a restaurant kitchen. There's a head chef, the router.

There are specialist cooks, the agents. You've got one for salads, you've got one for grills, you've got one for desserts, right? And there's a recipe book that gets updated every single night with new winning techniques, which is the learning loop. And there's also a memory system that never forgets anything, which is the vector memory.

And so when you send a task in, Ruffalo figures out what kind of task is it, which agents should handle it. Can multiple agents work on it at the same time? And has this been done before? What worked previously?

That's how it's self-learning right here. And then it can fire up the right combination and get the work done whilst you do something else. Now you might think, okay, what are these agents exactly? What is this and how does it work?

So an agent is just an AI worker with a specific job and Ruffalo comes with a hundred of them out of the box, right? So for example, you've got like content agents, you've got architecture agents, research agents, et cetera. You can actually get access to it directly online as well. So you can use a UI which is .io.

So it's ruv.io. So that's another option for using it. You can see Ruffalo is ready and installed right here. And then you can choose your default configuration preset.

So you can go with default, minimal, full or custom. So you can really change it a lot if you want to. So I'm going to go with default and then you can choose, okay, what type of agent swarm topology do you want, right? So do you want a hierarchical agent swarm?

Do you want a mesh, like peer-to-peer coordination, a hierarchical, adaptive, et cetera, right? So for complex projects, you probably want to go with hierarchical mesh, which is what we're going to do here. And you can say, okay, what are the max concurrent agents? You could put a hundred if you want to do, I'm just going to go with the default and then you can choose, okay, what do you want as a memory backend, right?

So it could be, for example, a hybrid with the agent DB and SQ lite, or agent DB or SQ lite or in-memory. I'm going to go with hybrid. And then you can enable better indexing. We can enable neural pattern learning, self-learning memory, embedding systems, et cetera.

And then we could choose our embedding model like this. Now it's going to get it set up. You might be wondering as well, what is a swarm? A swarm is a team of agents that organize themselves, right?

Which means that you don't need to assign every agent a task manually. You give the swarm a goal. The swarm figures out who does what. So three types of swarm architectures, hierarchical mesh and adaptive, right?

If you've never used swarms before, I would recommend that you start with the Hermes agent swarms. I've got it inside the air profile boardroom over here. We've got a full training guide and set up guide on exactly how to use it, how to set up, et cetera, but it's super powerful. We created like a full website set up, ready to go.

And it only took, it created a 50 page full website in the space of a couple of minutes. It was super powerful, but basically this means the system can handle like simple tasks and also massive complex tasks. So like I was saying, there's three types of swarm structures, right? You have hierarchical mesh and adaptive.

So the swarm can change its structure based on the task. You got mesh where every agent talks to every other agent and then you also have hierarchical, right? That's the difference here. So this setup is now complete.

As you can see, I'm just going to go over to core code, see how this works. And I'm going to send it the details of the setup that we've just done to get this working. And then you can see here, it's all set up, right? So now I'm going to say, okay, great.

Can you use the agent swarms feature from Ruffalo? We'll see how that goes. You also got a memory system too, right? So most AI tools have a huge flaw, which is they forget everything.

The moment you close the chat, Ruffalo doesn't. So it has a vector memory system, which is like a super fast filing cabinet with a genius librarian inside. So when you tell Ruffalo something, a preference, a result, a lesson learned, it stores it. And the next time a similar task comes up, it pulls up that memory and uses it, right?

So the search system Ruffalo uses is called HNSW, right? And it's up to 12,500 times faster than normal memory search methods, which means it just gets smarter the longer you use it. So you can see here an example, right? The Ruffalo swarm is up and running.

We've got this swarm right here. Here's the agents registered, et cetera, right? And so if I say, okay, go off and build an AI SEO website for a new blog, my AI profit boardroom community using an agent swarm, we'll see how that goes. Now you might say, okay, what tools does it connect to?

It's not just for Claude, right? So you can connect this to OpenAI, Gemini, Olama, et cetera, right? There's a lot of tools you can connect Ruffalo to, which is pretty cool. And also you've got the web UI as well.

So you don't even need to install it if you don't want to. You can go to this address and then just get access instead, right? Ruffalo has a web interface you can actually use right now. You want to see an example of what that looks like.

This is the example. I prefer to use it with Claude code directly, but you can just use it online if you want to. So inside there, you got access to like multimodal chat, 210 different tools, persistent memory, parallel tools, et cetera. Now you might say, okay, I need to be a developer to use it.

So actually the right belief is like, you can use it just on the web UI with zero coding. You just type in what you want. So you don't need to worry about that. Other people say AI agents make mistakes and I can't trust them.

That's fair enough. But Ruffalo actually has built-in validation and audit trails and self-correction baked in. So it's more careful than most human workers, to be honest. And if you've hired a virtual assistant online, you know what I'm talking about.

All right. And also some people say this is too expensive, but the thing is this is free and open source and you can just use it with Claude Code directly. So you don't need to worry about that. You can also see how it's working right here.

So it's registered the specialist agents and then it's kicking off the pipeline. Right. So it spawned the agents right here, as you can see, and it's beginning to build out that website. Some people say it would take months to set up.

As you've seen today, I've set it up in minutes, right? You just use the commands from GitHub and you're good to go. Do you give your AI their own emails? I have tried that in the past.

Honestly, I just don't like giving my agents access to email. I tend to find this. I only want to give it stuff that I'm comfortable with, right? And then actually use.

So I rarely give it access to my emails. The only time I would give agents access to my emails is if I'm using Gemini or something like that. It's literally designed for that task. Other people say AI forgets everything.

So it's useless for ongoing work. But Ruffalo's vector memory remembers across every session. So it gets smarter the more you use it. Other people say I need to micromanage every agent.

Actually, the right belief is you just give it a goal and Ruffalo's swarm figures out who does what, in what order, in what structure. So you check the output later, which is pretty cool. And you can see it working right here. So that's basically it.

That's how to use Ruffalo, how to get set up with it, how to use agent swarms. You can see an example of how it works. It saves all the details right here. So it says the site will be at this address.

We've got the list of the agents and their swarm status over here. So we've got an architect, a front end dev, an SEO specialist, and a reviewer. And then you can see their role and the status, right? And so, for example, these two are waiting for the architect and the reviewer is waiting for all, right?

So it's pretty easy to get Ruffalo working and set up. And then it will be available at this address once that's working directly. So pretty cool. So just to recap, Ruffalo is an AI orchestration platform that runs the topical code.

It gives you 100 specialized agents, each one built for a specific type of work. Agents work in swarms, self-organizing teams that handle tasks in parallel. It has persistent vector memory, remembering everything across sections to get it smart all the time. And you can try it right now, right?

It's free and open source. It's pretty cool. So if you want to get a full guide on how to use this stuff, feel free to get it at the AIprofitboarding.com. I'm going to add inside the classroom here a new tutorial called Ruffalo Agent Swarms.

And there you go. Right. So you got the full guide and the full step by step guide inside the AIprofitboarding right there. This actually comes with a full roadmap, a step by step tutorial on how to implement inside your business and everything you need to get it set up.

Also inside the AIprofitboarding, you can ask questions inside the community. You can jump on a weekly coaching course. You can get all of my best trainings inside the community, too. And then additionally, you can actually meet members inside your local city, ask questions whenever you want.

There's lots of cool stuff here. What we actually do as well is we keep a changelog of all the improvements we make every day so that you can see how we're improving, how we're developing over time. Right. So you can see that we're just adding new improvements every single day and just rapidly evolving and improving as a community based on people's feedback.

So feel free to get that link in the description or just go to the AIprofitboarding.com. Can it work with the Kanban? No. If you want to get agent swarms working with Kanbans, check out my training on Hermes Kanban here.

Right. This is super powerful. So if you get that right there, that can show you how to set up Hermes agent swarms with Kanban using Hermes Workspace. Super powerful.

HUD is Hermes. So that's a totally different topic. Hermes HUD doesn't really work with Kanban. It's more for like just managing your agent and the back end of it, I would say.

How does Hermes HUD manage token usage? So you can see the back end and like the costs on the back end. So that's how you manage it with Hermes HUD. Hi, sir.

If I want to start now, what tools can I start with? I want to start with better agents. Yeah, so I would go with Hermes, right? Out of all the tools, all the agents in the world, Hermes is by far the most powerful.

We've got a full tutorial and guide on it right here. How to set up, how to use it, et cetera. But yeah, I would go with Hermes. Like super powerful, really easy to use.

Super smooth, way easier than using OpenCore. Go with that.