DeepSeek Harness: FREE 1 Hour Course!

This is the full DeepSeek harness course, and by the end of this video, you'll have a free AI employee working at your machine 24-7. DeepSeek just dropped a brand new tool, it's completely free, and the thing turns your AI from a chatbot into a real worker that reads your files, writes your drafts, and does your actual jobs for you whilst you're at the gym. It's one of the best coding harnesses out there, one of the fastest growing GitHub projects ever, and I want to take you through the three levels to learn this bad boy. So level one, you'll have it running in under five minutes.

Level two, I'll show you how to build your own custom tools and automations, apps, websites, just by typing one single sentence. And there's a hidden setting inside this level that almost nobody will ever find, which I'll reveal later. Level number three is the end game. Your own team of AI agents all working together whilst you sleep with DeepSeek harness.

So stick with me until the end, and I promise you'll walk away with your own free AI employee, no matter who you are. Let's get into it. Now, quick facts before we start, so you know what we're dealing with here. DeepSeek harness came out on August 13th as a developer preview.

It's completely free, open source, MIT licensed, which means you can download it, you can change it, you can even build a product on top of it without asking for anyone's permission. And also it blew past 115,000 stars on GitHub, which puts it among the fastest growing AI projects ever recorded. There are already over 300 community plugins for it, sorted into categories like knowledge, media, workflows, integrations on a tool that's just days old. The project is led by Tianyi Cui, who joined DeepSeek in March 2026 from Jane Street, one of the sharpest technical firms in the world.

And a team of 19 people took this from a standing start in May to a public release in August. Also, some people will be watching this and be like, well, the harness is free, but the brain isn't free because you have to pay for the API. I'll show you a way around that to get a free brain plugged in later in this course. Let's get into it.

So let's get into level one, which is installing it, understanding it and putting it to work. First of all, one minute on what a harness actually is. If you get this one idea, the entire rest of this course clicks into place. So don't skip it, even if you think you know.

So an AI model is a brain, just a brain, and it can think, but it cannot touch anything. It can't open your files, it can't remember what you told it last Tuesday, it can't use a browser or run a program or save its work. On its own, a model is a very smart voice in a jar. The harness is everything wrapped around that brain.

So it's the hands, the memory, the workspace, the rules about what it's allowed to touch. When people say AI agent, what they actually mean is model plus harness, brain plus body. So for example, Claude Code is a harness around Claude. Hermes agent is a harness around any agent model that you plug into it.

The model gets the credit, but the harness is where the work actually happens and harnesses matter more than most people think. So a top model might be five or 10% better than the one below it. A good harness makes the same model dramatically more useful because it's a harness that decides what the model can see, what it can do and how it recovers when something goes wrong. The industry spent two years obsessing over brains, but now it's really moved to bodies.

So DeepSeek harness is DeepSeek's body. Until now, the serious harnesses were mostly closed products. So for example, you had Claude Code, you had Codex and you could switch the brains around those in them. But DeepSeek harness is really designed to be open source and easy to switch around the models and pretty much switch everything around it.

So on August 13th, DeepSeek published a direct alternative to that whole ecosystem of Claude Code and Codex and Hermes and everything else. And this is MIT licensed, it's on GitHub for everyone. And the same lab that shook the model world in early 2025 just did the same thing to the harness world this year. And that's a context.

Now let's actually get it on your machine. So step number one, Node.js. This is the only thing you need installed first. Node.js is a free program that lets this kind of software run your computer.

You can go to Node.js.org, download the version for your system and click through the installer like any normal program. Next, next, finish. Two minutes. If you already have it, skip ahead.

Step number two is you just want to open this up inside your terminal. So on a Mac, you can press command plus space, type terminal, hit enter. On Windows, check start, type PowerShell, hit enter. I know the terminal, not the nicest place.

I know it's like a bit intimidating for some people, non-technical. But you're just going to type one line. And after today, you may never think about the terminal again because this tool doesn't actually live there. And step number three is you just type this command and hit enter to install it.

That's it. Literally this command, hit enter inside your terminal, just like you can see right here. And you are good to go, my friends. And once you've set it up, it's going to look something like this.

And that's the whole install. So that command downloads DeepSea Karnas and starts it. The first run takes a minute or two whilst it pulls everything down. And when it's ready, it gives you a local address.

By default, it's your machine port, which is 3080. And it opens a page in your normal web browser. I actually like this about DeepSea. I think it's quite a smart choice, which is that it doesn't keep you in the terminal.

It runs almost like a normal web page. So you have a chat window, sidebar, buttons, folder picker. It looks like software you already know how to use. So you can see ours running over here.

Almost every other Serious Karnas lives inside the black terminal window, which quietly filters out everyone who isn't an engineer. You can, for example, use apps like Claude. But again, like it's another app, it's another button you have to click outside of your ecosystem. Because most people are just directly inside the browser for most of the work they do.

With this system, it's just like inside one page. And you can just run it and have it ready to go. Or you can plug it into an agent operating system like you see right here. And it's way easier to use.

I think DeepSeek shipped this web first on purpose, right? Because the page runs locally on your machine, but it's not a website someone else hosts, it's yours. And then step number four is you want to connect a brain. So the first time the harness opens, as you can see right here, it asks for a DeepSeek API key.

An API key is just a password that connects the harness to a model. You can go to platform.deepseek.com, sign up, click API keys, create one, copy it, paste it into the harness, and that's done. That connects you to DeepSeek's models, including V4 Pro, their flagship built heavily for agentic work, which shipped the very same day as the harness. So the brain and the body launched together.

They were built for each other. Now, you can also change the economics of this as well. Because inside your settings, under models, there's an add provider button, which can connect OpenRouter. So you can use a free brain inside this.

So, for example, if you connect OpenRouter, that uses cheap and free models. If you use Olamo with the DeepSeek harness, you can run free local models with this, if you've got a decent setup. You could also, for example, find a free model of OpenRouter plug-in. Or you could use DeepSeek for V4 Pro, which is their flagship model.

And that's still pretty cheap, but it's not free. And then you can add anything else. So you can just go inside your settings over here, and then we can go to models, and then we can add a provider, like so. We can flick down the list, and we could be like, okay, I want to use OpenCode.

Now, here's what's pretty cool about OpenCode. OpenCode has free models. And guess what one of its free models is? DeepSeek V4 Flash.

So you can actually get a free provider by OpenCode, plug in the API key from OpenCode, and then, as I promised earlier, that's how you can use free models with it. And then the harness is free, and the brain is free, and you've got a full free system ready to go. Now, you can also give it a workspace. So in the interface, if you click the folder icon, you can add a workspace.

This is pretty useful, so let me explain it properly. Basically, the workspace is the room that your agent works in. So you see how we can select a folder that we want it to run in, and we can add a workspace like this, and then just select a local setup. Well, basically, this is the room that your agent works in.

So it can read every file in that folder, it can create files in that folder, it can change the files in that folder, and it can't touch anything outside it. So you don't want to point it at your whole computer, because you want to make one clean folder, right? So don't point it at your whole desktop. I have over, like, 10 years of clutter on it.

You want to make one clean folder, call it agent workspace or whatever you like, and put copies of the files you want it working on inside. So that could be, like, your draft content, your research notes, your meeting notes, whatever your actual work is. So it's one room with a door you control. And this keeps things safe, and also just as important, it also keeps the agent focused on one single place.

Because you don't want, like, an agent digging through thousands of irrelevant files. That would actually cost a lot of tokens, cost you time, and also it would be super slow and unfocused. So instead, create the workspace. Make it nice and clean.

One clean folder. And then step number six is your first real task. And you don't want to just, you know, write a poem with this or something like that, right? You want a real piece of your actual work.

Because that's the only way you can really start getting the most out of this. So look at what you spend your time on, look at something that you spend on every single day or a cool idea that you have, and then you can start building with this directly, right? And the way that you use DeepSeek, as you can see right here with the harness, is you just write in plain English what you want. Exactly the way you'd brief a smart new assistant who is working for you.

So here's a real example. We could say, read all the notes in this folder. Write a one-page summary of the main points as a new file called summary.md. Use simple language and short bullet points.

And then you can see, for example, here, that it comes back with all the responses. And this is a harness that is agentic. So it can run on long-horizon tasks. It can do stuff on your computer.

It's not just a chat window. It actually thinks, it lists the files in the folder to see what's there. And it opens up one by one, and it works through the content. And then it can actually write a new file into your workspace.

So the great thing about this is you can use it for building apps, for building tools, for building websites, for building, for example, games even, right? A harness does work in the place the work lives, and the output is something real that you can open, send, or build. So, let me give you the pattern for writing good tasks because the quality of what you get back is mostly decided by the quality of what you type in. So, there are three parts to this.

What to look at. So, for example, like, read the notes in this folder. What to produce. So, for example, a one-page summary as a new file called SummaryMD and what rules to follow.

So, for example, simple language or short bullet points, right? Look at. Produce rules. That's a whole skill.

And you don't need, like, magical prompts. You just need clear briefs. Like, you just talk to this the way you would message a friend. So, if we wanted to create a new session here, for example, we can just click on new session and we say, okay, build me a beautiful website for my SEO agency and use my local Obsidian folder to come up with context so that you understand what my SEO agency is about.

And then we can plug that in and then we can select the folder we want it to work from. So, for example, here we could go inside Obsidian Vault. We can plug that in. We can click choose and then we can hit enter like so.

And that's going to begin to work its magic. And that's a whole skill. You don't need anything fancy here. If your instructions would confuse a human assistant though, they'll confuse the agent.

So, you want to make this clear, easy to understand, give you access to relevant local files and just make it very easy to use. And you can use this for, like, content creation. You can see how fast it runs as well. So, it runs really quickly.

We can also see a full session log over here. We can add new sessions. We can see multiple projects that we're working from, what's ungrouped as well. And then we have, like, our plugins over here too.

So, plugins, for example, we've got, like, web search and we can allow it to search the web. We can set up agent loops. So, how many parallel tool calls it runs. We have the shell.

So, shell command timeout as well. And plugin list. So, we can actually add new plugins that can access tools from outside. So, for example, over here we have, like, web enabled, which means that it can actually search web using a plugin that we've installed over here, which is pretty nice.

We'll come on to more stuff like that and, like, you know, more complex things on that as you can see. Now, obviously, you'll see as we use it, it's like, who's this site's primary audience? We can click on this, for example. We can say, what's the single main action that we want?

Book a strategy session. How do we want it to look? So, modern minimal sass. And then also one landing page.

So, you can see that it's going to give you options to select. So, where should I build the site files, for example? We can select document folder, hit submit, and now it is deep diving through everything that we're creating locally here. Pretty nice.

Now, at this point, you might be wondering, okay, you know, where do you start with this? How do you get the most value out of it? You know, all these tools, all these different apps, I know a lot of people who are not focused and not getting the know out of these systems and that's why I don't recommend to you. So, the main thing here is that you give this harness real jobs from day one.

And if you're not sure what real jobs are, look at what you're working on and then go from there. So, for example, if you make content, drop your notes in the workspace and help it create content for you. You know, that could be, for example, if you do SEO, get it to automate your SEO for you. Whatever you work on day to day, that's what you should be giving to DeepSea Karnas.

Don't use it to just vibe code a snake game or something. So, the main point here is that you make it actionable. You make it actually useful and you can just focus on one automation per week. So, for example, for me, what I like to do is I'll measure my time, I'll measure what I work on day to day and I'll put that as a list and then I'll be like, right, okay, well, out of all these tasks, this will be the easiest for me to automate next with something like DeepSea Karnas.

So, if I'm looking at this, I'm like, right, okay, well, each day I create content for my website. So, let's make sure we automate on a schedule, creating content on my website using DeepSea Karnas instead of me creating it daily. And that way, we actually turn it into something useful instead of just going after like random use cases that don't make any sense. Now, let's also talk about a useful feature that's quite unique to DeepSea and it's called the trajectory view, right?

The trajectory view, as you can see right here. And the way that this works is quite interesting. So, next to your session, there's a trajectory panel. You can click it and what you get is every single step the agent took laid out panel, graphically clickable in order.

So, if we have a look, for example, here, we can take a look at the trajectory. So, these are two different versions. The chat is just plain text. You just talk to it in plain text.

The trajectory is you can see what's going on and you can see it's clickable in order. So, for example, we can click on this and we can see what's happening and we've got the summary, the preview and everything else. Now, also, as you're building with this, you're going to see that it asks you to reject or approve or allow stuff. So, as we're going along, we can allow it if we set up the permissions.

If you don't want to have to click through that every single time, you can click on full access and then enable full access. Only do that if you're not comfortable with it, don't do it. Makes sense. Now, as we go through the trajectory, basically, this works in order.

So, what you can see here is like the system prompt it was actually given. So, if we scroll right to the top inside this trajectory view, we can see the initial system prompt. And basically, this is what it was given at the start. So, this is your message, the context it loaded, its actual thinking, the reasoning visible in full, every tool it called, the exact command run it ran and the exact payload it sent, the exact results that came back and how long each step took.

Also, it shows you, for example, how many turns the whole job needed. So, you can click any single step and read the raw detail underneath it. Nothing is like hidden. So, why does this matter so much in practice?

Beyond it like being impressive and useful and looking super cool. So, basically, AI agents make mistakes, all of them. Anyone who tells you otherwise is basically, you know, they haven't tried AI agents. They break all the time.

Agents break all the time. They are easily misled, easily distracted. They're not at the point where they can just build and automate anything without having an understanding of how it works. So, the difference between agents you can trust with real work and agents you can't is whether you can see why a mistake has happened.

So, with most tools, a failed job gives you nothing. The output is wrong, good, luck, understanding why. With the trajectory view, you can scroll to the exact step where things went sideways. So, you can look through here and you'll be like, ah, it looked in the wrong folder.

Ah, I misread that one line inside my brief. Ah, the file it needed was empty. So, you can fix that one specific thing and then run it again. So, we can have a look, for example, here we can see what it did, what it worked on.

And then if we want to, we can basically take that and then run it again based on where it's broken. This is also the complete opposite of the direction that a lot of closed labs have taken where basically the model's thinking, it's kind of like hidden from you. You know, you can see what's going on, but you will never get as much detail as this. Like, this is super, super useful compared to a lot of steps.

So, DeepSeq made every step visible. It's kind of like a CAD core design, like principle. So, once you've managed an agent, you get the full vision, you can see everything that's going on, and it basically gives you a lot of visibility and understanding. Now, also, what you can do, and this is another level, is you can get the session log.

So, you can actually export the full session as a zip file. And inside is a complete log of every step. So, you can export any session as a complete log. And with this, you can see every step, every call, every result in a standard file format.

So, you can save it, and you have a permanent record of exactly what the agent did. If you later hand agent work to a virtual assistant or a team member, that log can help you check the work without having to redo the work. So, you review the trajectory the way a manager reviews a report. Last thing in level one, by the way, and this is still level one here, is this version is 0.1.

And what I mean by that is basically DeepSeq have only just released this. This is 0.1. Basically, it's 10% of the full version. And so, it's a developer preview.

And what this means, essentially, is be expected for things to be a little bit more. Anything open source doesn't always work first time. So, it's pretty cool that we can see the chat. We can see the trajectory.

It can build awesome stuff. As you've already seen, it's pretty easy to use. But bear in mind, there's going to be things that break. There's going to be things that move around in between versions and updates.

A plugin that works today might break next month and get fixed a month after. So, if you wire your entire operation to it this week and treat it as finished software, something's going to break. And I just want to be honest with you right now, because you can start doing real jobs with it. But basically, you want to structure it around a system.

And I'll show you that later inside level three. And it's one of the strongest ways to use it. So, you want to build this into a system, which we'll come on to later. But that's essentially level one complete.

Well done for making it this far. So, one key connects the brain. One folder is the room it works in. Plain English briefs with a look at, produce, rules, system, full trajectory, vision on every step.

Session logs, so you can keep and check everything. And basically, if you stopped right now, you would have a free working, fully inspectable AI agent doing real tasks on your machine today. But you'd only be using 20% of the full power of DeepSea. The other 80% is the part that made me make this course.

So, that's level one. Now, quick pause here. Before we get into the rest of it, everything in this course, we run live inside the AirPuffer boardroom. And the DeepSea harness is on the daily tutorial schedule right now.

The exact install from level one, connected free models for open routers, the agent work costs nothing. And also, just like how to get the most out of AI automation. So, if you want really practical, useful stuff like this, if you want weekly coaching calls, if you want to connect with amazing people, feel free to check out the AI Profit Boardroom. Link in the comments description.

or go to theairprofitable.com. Every single day I answer your questions, I create personalized video tutorials for you, I have a classroom with all these amazing courses, you can jump a weekly coaching course right here, you can meet people in your local area who are building out with AI agents and stuff like this, and it's just a great place to learn about Deep Seek, AI agents, you can meet cool people, and it's a fantastic community. So with that, let's get into level two. Level two, my friends.

The plugin system, the four modes, creator mode, the hidden config file, how the model works, and the one line in the docs that quietly connects this tool to every other agent that you own. So let's start with the headline of Deep Seek's own GitHub, which is three words. Everything is a plugin. Okay, so most software companies say something like that and mean like we have an app store where you can add extras.

Deep Seek means it literally, and the difference is the whole story of this tool. So the sidebar and interface is a plugin. The tool the agent uses to act, the plugins. The system prompt, the standard instructions, the agent starts with a plugin.

The sandbox, where it works, a plugin. Even the agent loop itself, the core cycle, think, act, check, repeat is a plugin. So every single organ of this machine can be switched off, swapped out, or replaced with a better one without breaking the rest. And here's the picture to hold in your mind, my friends.

So every other agent product is like a sealed car. You can change your radio station. You can maybe add like a phone mount to it, but the engine is the engine, the seats are the seats. And if the company that built it makes a bad choice, you're stuck with it.

That's it. But Deep Seek harnesses car where the engine, the wheels, the seats, and the steering wheel all pop out and click back in and the car keeps driving whilst you do it. So when someone else builds like a better memory system next month, you swap that one part in. When a better browsing tool comes out and gets shipped, same thing.

So you don't wait for the company. You don't migrate to a new product. You upgrade one piece and keep moving. And someone will build better parts, right?

That's what the community is all about. So let me give you some examples. Within hours of this launching, and again, I only launched like this week, Deep Seek harness, it's already happening at a pace I've never seen before, right? So for example, within hours, the community chipped a desktop version and a curated plugin list.

There are over 300 community plugins indexed right now from the official plugin tag, sorted into categories. You've got knowledge, media, workflow integrations. Deep Seek even tells plugin builders to tag their repositories with dsh-plugin so everyone can find them. And when a free tool grows an ecosystem in its first days, that really tells you where the builders believe the future is, right?

A lot of developers are switching over to Deep Seek harness because it's got that open source vibe. It's so easy to share with them. So under the hood, this whole design runs on something called Cordis. And Deep Seek did something labs almost never do for a product like this.

They actually published an actual research paper explaining the design. Now, if you haven't seen that already, you can actually find it on the GitHub page, right? So they have a full like paper on Cordis and how it works as a paradigm. And essentially, you don't need to read it.

It is kind of interesting, but it's very, very heavy maths. But the core idea matters and it's pretty simple to say, right? So every action in the system can be cleanly reversed. If you add a piece and there's a guaranteed clean way to remove that piece with no leftover mess and no side effects and everything else, well, that's the reason everything is a plugin actually works here, right?

So instead of collapsing into chaos, which is what usually happens when software gets this modular, the swapping is pretty safe by design mathematically, not held together with like tape and hope. There's also very Jane Street. So if you know that world, building the boring and rigorous foundations first, so the exciting stuff on top doesn't fall over. That's what it's all about.

It's just about building good foundations inside this harness. And again, like I said, it's very unique. Now you might be wondering, okay, what are the four modes? Because this is the control panel people click past on day one and never revisit.

So if we have a look, for example, here, you can see here that we've got this standard mode right here at the top, right? So we can switch through this when we start a new session. So we've got standard mode, code mode, minimal mode, and create a mode. So in settings, there's an agent preset option with four choices and each one is just a really different amount of power handed to the agent.

Standard mode is home base, balanced capabilities, good judgment about tools, right? For everyday business tasks, the summaries, the drafts, the research jobs from level one, you'll spend most of your time here to be honest. Code mode, right? That actually unlocks the heavier building powers for bigger technical jobs.

So if you're having to construct something complex over many steps, this gives it more room to work. Most non-technical users will rarely use it. And that's fine. Knowing you don't need a mode is probably as useful as knowing you do.

And then for example, you have a minimal mode. And what this does is actually strips the agent down to almost nothing. It's very light, it's very fast, fewer moving parts, perfect for quick, simple questions where you don't need it opening files or running tools for you. Think of it as a difference between calling a whole team into a meeting and just asking the person at the next desk.

And then there's create a mode. So create a mode exists for exactly one job and it's a feature I teased at the top of this course. So building new plugins by prompting. You can describe a tool that you wish existed in one plain English sentence and the harness builds that tool live inside your session and actually switches it on.

So let me walk you through this slowly because the first time you see it, it doesn't really feel real. But if you start a new session, you can pick create a mode. So let's try this out. So we're going to click on create a mode over here like so.

We'll select that. Make sure you click on a new session. So you click the plus sign to start a new session and then you're going to select create a mode. And from there, you type a sentence describing something that you want to exist.

So let's make it concrete. In this example, we actually said add a small panel in the bottom corner that shows my three tasks for today and let me tick them off. So you can now send that as you can see right here. And then we can click on full access and we'll just allow that as you can see right here.

And basically, the moment you send that, the harness automatically loads its own plug-in development skill. So you can watch it do this in the trajectory view. So if we go to the trajectory view that we were talking about before, you can see here that it's beginning to use this system right here. And so if you look at this example, this is a sped up version.

Basically, when we've asked for the system, it's got the daily panel over here. So we asked it to add a small panel. This is again, sped up just so you can see it in action. But the main thing here is you can build your own workspace.

How amazing is that? Imagine you can go into Claude and you're like, mate, I want this to look like that. I want this to look like that. I want to build my mission control.

You can't do that inside Claude. It's closed. It's not editable. It doesn't swap in and out.

But Cordis is unbelievable for doing this. I've never seen any other tool that can do that. And I don't want to hype it up or tell you how amazing it is without you trying it and then you try it and maybe it's not as great as it sounds, but it is impressive. Like, I've never seen that before.

So what this allows you to do is basically build something in real time inside the plugin workspace and just add new cool stuff. So if you're building out and you're like, do you know what? I wish I had a goal section. I wish I had a journal section.

I wish we could do this. I wish we could do that. You can do it, right? It's kind of like a genie.

So it starts building. A minute or so later, it presents the finished plugin and asks you to approve it. One button, you click approve, and the panel appears on your screen. So you can now tick off tasks.

You didn't write a line of code here, by the way. You didn't download anything. You didn't wait for a developer. You just described what you want and the thing came into existence in front of you.

How amazing is that? Now, if you don't like it, you can edit that as well. So you could say, okay, make it smaller or move it left, or I want different colors or keep it forever. So you literally describe what you want.

And so it's a very editable and sort of a flexible harness that, again, I've never really seen before. So it's like a conversation. You can go back and forth and change it exactly how you want to mold it in the vision that you have for it. Now, you can also stop the plugin and then it can basically end instantly.

So you can actually switch off anything that you have as well, and you can start it and it's back. So by default, it just lives inside that session, but you can also persist it as well. So it's permanently part of your harness. So you see how it's now defining the plugin.

Again, this is slow motion mode versus the demo that I showed you a second ago, but that's basically how it works. And then if you build something, you can keep it forever. So you see how it's just built out this task panel now and we can approve it and boom, look at that. We've just built out a list of goals that we want.

And then if we want, we can make that persistent forever, or we can just delete it and we can add tasks. We can edit it. We get to get off and that's ready to go. Unbelievable.

Really cool. I just love how flexible that is. You're never going to get that with a closed model, but this is the power of open source. It's just, it's designed for what you want.

Also, when you think about this, it's like, okay, well, who are they targeting? I think they're targeting people, particularly developers who use stuff like Cloud Code day to day, but they're like, I want it to be more my style. I want it to be customized to me. And so this is, this is something that you can do.

Anyway, you can basically build stuff, keep it forever. And if you think about what this actually changes, like every piece of software you have ever used was somebody else's guess about what you need, but the menus, the buttons, the panels, it was all designed for an imaginary average user who isn't you. Creator mode flips that relationship. So the interface becomes clay.

You want a panel that tracks your content published from idea to publish, describe it. You want a button that kicks off your morning routine with one click, describe it, right? A little dashboard showing the last five files your agents produce, you just describe it. gap between I wish this tool did x and the tool doing x collapses to one sentence and one approval click and I've been doing this for a while and I still find that slightly hard to believe every time it works.

Now the buried setting that I promised you at the start, this one's for when you want control even creator mode doesn't give you the power to switch off pieces of the product so on your machine there's a dsh folder all right so dsh like so and inside its profiles then web and in there sits a file called cordis.patch.yaml you can open it up in any text editor notepad is fine it's a short simple list now here's what you can do with it you can actually write the id of any plugin in the system and remember everything is a plugin including core interface pieces like the cyber and you can add like one indented line underneath that says disabled true right then you save the file reload the harness in your browser and that piece is gone so you can edit basically the config file inside your dsh and change it and mold it however you want or you can get the agent to do it for you so that's what they mean by like editing the harness you can basically change everything you want to and then you see how we've now got the cordis plugin section over here and we can delete stuff we can stop it we can play it run it change it however we want to and it's all ready to go inside the plugin section over there which is pretty nice and if we go to plugin list you can see all these other plugins and the plugin configuration too now you can also view and see how each of these presets is configured so if we open up this configuration file like so we can click on that and open up and edit it but then we can switch between these different modes and we can also add and draft a custom preset so if we go to settings here agent presets you've got minimal creator standard and code but you could also draft a custom preset as well if you wanted to add like a totally different mode that again is fully customizable and that's down to your creativity like for me i'm not really going to mess around with that i don't think it's going to help me too much but if you wanted to add like a new mode that doesn't fit within one of these four boxes you could easily do that as well now there's also these websites that are popping up with individual new plugins that you can plug into these systems as well right so for example if we look there's a website called deepseek-code.com forward slash plugins and what you can actually do is browse like all the plugins have been built inside this system so for example they've got like plugins for developers which is like quite technical but then you also have like these agent skills so you see here portable agent skills for building publishing and remixing games and that sort of thing and then we can view it on github and we can actually ask for example our agent to pull that in and install it to github uh to our agent now when you're installing anything from github you need to be careful check it yourself get one of your agents to check it out as well make sure it's legit and you're happy with it but basically if you wanted to install something like this you could go over to your plugins section inside the settings here and then you can search for plugins and you can install new plugins or you can actually go inside the new session here and then just say okay install this plugin and then you can pull in the plugins from other people's stuff so you see how it creates almost like an app store for deepseek harnesses where you can share everything in the world you know everything is a plugin every single part every single piece you can swap it in swap it out so that concludes level two of this whole course before we actually get into level three there's just one final bit i actually want to mention about level two and it's quite a line in the github repo that i think matters more than the whole launch announcement so the docs say that deepseek harness supports mcp supports the agent client protocol and reads agents md and claude md files so let me explain that in plain english all that means is it's the bridge to level three because mcp is a standard plug socket for tools is how any agent can connect any tool without custom wiring for each pair so the agent client protocol is a standard way for agent software to talk to the apps around it and agents md and claude md these are the big ones are instruction files so little text files you leave in a folder then tell that agent about your business so this could be your rules your writing voice your projects what to always do what to never do where things live claude code read some open claw read some hermes read some and now deepseek read some as well and reads the exact same files so if you don't have these files yet start one today because it's the highest um potential thing you can actually do and it shares across all your agents it takes about 20 minutes you can make a file called agents md in your workspace folder or you can just ask your agents to set that up for you you don't have to do it yourself and then you can write and play in english like what your business does who your customers are how you like things written short sentences no jargon for example whatever your rules are what your current projects are and anything about the agent and what it should never do right and that's it it's a briefing document but the one you'd hand to a new hire and then from then on every agent that opens that folder starts every task already knowing your business i also really like for example something called obsidian obsidian is a fantastic way to plug in context on a more advanced level so what this has is like a full database as you can see right here with all my notes about everything about them so for example if i am using deepseek i can plug that in and this is more part of the agent operating system but i can plug in all my context and all my notes and memories and everything about me into one system and then that will get pulled in to our deepseek harness and so that makes everything easier because then you know whatever new harness comes out tomorrow that could be even better we can just plug this system in and make it even better and if you look at what deepseek chose here they could have had invented like their own special instruction format and built a walled garden but they did the exact opposite so they built the harness to speak the languages the existing agent world already speaks like claude md the format the biggest closed source rival created a challenger that builds for compatibility instead of locking it's just hoping for adaptability right they're hoping that developers come along see that and they're like oh we need that right people like you watch this and you're like oh i love the plugin system let's build that in so this leads up to a very special question which is if every agent on your machine reads the same instruction files works in the same folders speaks the same protocols why are you running them separately at all and that brings us to level three the agent operating system with the deepseek harness inside it so let's talk first about what this problem actually solves and i hear it from my ai profit boarding members all the time so if you've followed ai for the past year you've probably collected tools and subscriptions so for example claude code maybe hermes maybe open claw now deepseek harness each one arrived with a wave of excitement each one demanded setup time and somewhere along the way a feeling crept in there are too many of these too many tools too many apps i can't keep up every month there's a new one and i'm falling further behind some of you are feeling it about this very video so i want you to take that feeling apart properly because it's built on a picture that's wrong the picture in your head might be a treadmill every new tool is another thing you must master so more tools means more falling behind but mastery of individual tools was never the skill the skill is running one system that new tools plug into and when you have the system a new tool isn't homework it's a free upgrade delivered to your door so you don't keep up with the tools your system absorbs them and here's what should give you real confidence in this approach right deepseek have focused their entire ecosystem on this whole system right they they have literally built out a harness where you can swap everything in and out and that's basically what an agentic operating system as well it's the same philosophy so everything is a plug-in inside one tool the agent os same idea one level up every tool is a plug-in inside one system and the biggest open lab in the world just validated the architecture so let me show you what it looks like in practice because this is exactly how i work every single day across my business so we've got one dashboard i call the mission control and every agent sits side by side each with his own job out what it's best out so if we pull this up for example this is my genetic operating system and what we can do here is we can see that we've got these different workflows like the seo workflow we've got a video agent we have the loop system we have notebook lm plugged in here we've got a kanban board memory systems etc right everything is in one place one click away including deepseek and deepseek harness itself beautiful so that makes everything much easier to manage and that way we can have all of our agents and everything inside one system and i can rebuild any piece of it and the thing that makes this a system instead of a pile of apps is the shared memory with obsidian as i mentioned before so every agent reads and writes to one obsidian ball if you haven't met obsidian it's just a folder of plain text notes on your computer that's genuinely all it is and that plainness is the superpower because plain text files in a folder are the one thing every single agent can read and write no special format no company's database it's just like one agent learns and then every agent knows because they're all looking at the same folder so here's how to structure yours and how to make sure that actually works so i like to use daily projects research and outputs that's the way that i organize this whole system so it looks beautiful like this but it's just like organized into four steps daily projects research output and it's just a bunch of folders with one instruction folder the reason i do that is because complicated vaults don't work very well but a handful of folders you know one for daily notes what happened what's in progress one for projects one note per active project and for example what its current state is one for research everything the agents gather and one for output drafts and finish work and sitting on top is your agents md instructions from level two so every agent that opens a vault briefed automatically, and that's the whole architecture. It's just folders and text files, pretty simple stuff. If you can organize a filing cabinet, you can build this. So if we come back to DeepSeek and DeepSeek Harness here, and we go back to the ideas it was pulling in before on our sessions, you can see that we've got everything pulled in for our website.

So this is the whole website that is pulled in, and it's all personalized to me, right? So it knows my brand and everything else. Why does DeepSeek Harness have so much information about me from day one? Because it's pulling in all the context from Obsidian.

So it knows what we work on, it knows what we do, it knows who we work with, it has all of our FAQs, it understands everything about me and my business from day one, because we just get that memory and we plug it into our agents. And we can do the same with Hermes agent, we can do it with all our agents. It's an amazing system. Game-changing once you understand it as well.

So this is the whole system, and then for example, if I'm doing something in Claude, I can go over here. If I wanna go back to DeepSeek, I can go over here, right? Now you might be wondering, okay, how do you build out an agentic operating system like this? Well, I mean, if you want mine, you can get it inside the AeroProfit boardroom, so it's already set up and ready to go.

If you want to just build it like this, you can literally just create a new session, start a new folder, so we can create, add a workspace here, and we can say mission control, like so, and then we can open up, and then we can say, okay, create a beautiful mission control dashboard for DeepSeek Harness and also my Obsidian memory locally. And then you can start with the tools you use. You know, if you're using Claude code day to day and you want that next to Hermes, you just go with the same. And then from here, we can select this.

Now, also you can switch between the models here. One thing to note, so for example, we can select V4 Pro, which is the most powerful frontier level AI that DeepSeek has, whereas for example, DeepSeek V4 Flash, not as powerful, but great for volume. So if you're like, okay, when do I use Pro? When do I use Flash?

Flash is great for volume, it's pretty cheap, but not that smart. DeepSeek V4 Pro, fantastic for building things. So if we're building something that's like the agentic operating system, we can just click enter like so, and that can begin to build out. Now this is basically like a dashboard, you know, a mission control.

So it's going to code out and then it's going to start building it locally, right? So you can see here, it says mission control folder is empty, a fresh build. Let me read your canonical memory file and search for what DeepSeek actually is. And then it's beginning to build this out.

And as you can see right here, it's beginning to build this in. Now you can also see what it's doing step-by-step. So it's got like context injection, and then it's got, for example, thinking, and then it's running terminal commands. So bash is like terminal commands, and it's just running through everything step-by-step.

We can also see what it's working on locally. So what it's doing here is actually reading through our obsidian vault, understanding how our existing mission control is used to start building that as well. And then it's even, for example, checking out GoldieBench, it's asking for permission locally as well, and everything we're building, and we're good to go on that. So that might code out for like 10, 15 minutes, but basically what we're getting there is a full, beautiful system that we can build with the mission control.

And you can do that with any agent. You don't have to do it with DeepSeek. DeepSeek I'm just using, because this course is about DeepSeek, but you could build out a mission control using obsidian. You could use it with Claude or whatever you want.

And then you can see here, it's like I've mapped out the environment. So we've got the mission control folder, everything else, frame. And it's like, okay, what do you want me to actually do here? So what should it be for?

This is for DeepSeek harness. And then we'll get a full control room and it'll begin to build out. As you can see right here, which is pretty nice. And anytime you want to stop this, we can go with it as well.

You're also going to see here that as this bottom right hand corner, the context used. So it shows how many tokens we use, the system prompt, the tools and messages. So it's broken down how much context, everything is taken up. So messages is actually taking up the most tokens and context right here.

And then we can see, okay, how many steps we've used, the LLMs, everything else, right? Cache here as well. So that's based on like basically reducing the amount of tokens we use by rereading the existing context we already have. And it's beginning to build this out.

And so we can have that working in the background. We can also open up other tabs of our mission control and I'll walk you through this. So over here, we have the mission control dashboard. We have all these different open source projects like Hermes, sorry, Paperclip is a system where we have basically a team, like an organizational chart of all our agents working together.

We have the AI agent mastermind where we can have DeepSeek and all our agents speaking inside a group chat, like so. We have a pipeline system here where we can go from idea to built app. Then we have, for example, like Claude with the workspace, so everything is saved and we can preview it for later. We have artifacts that we can preview, ultra code where we can have teams of sub-agents working with Claude, and the ANSI, like an agent system as well.

Then we have Hermes here. So you see how each tab is basically like a new layer. So for DeepSeek, you can have different versions of DeepSeek in the harness or different custom workflows that you actually prefer to use. And it's all slotted into one system.

And if you're like, this is going too fast for me or that sort of thing, don't worry. This is level three. But if you want like a full breakdown, we've actually got a one hour tutorial inside the Air Profit Boarding on exactly how we build the operating system, what each part does, and how it works step-by-step as well. You'll also see that Claude is, sorry, DeepSeek is pulling in the skills.

So it's got these skills here, like front-end design, hallmark is for reducing the amount of tokens that she uses. And then we've got front-end design as well right here. And it's using those to create better quality designs for our mission control dashboard. So over here, we have Hermes.

We've got, you know, like the Hermes bot, new feature that just dropped. So we can kind of use Hermes like a WhatsApp where we have different chats. We've got Hermes Apollo, it's a voice agent right here. And this just gives you a feel and a vibe for how much more fun AI can be when you build out these custom workflows, custom UIs, that look beautiful inside one system.

So that's what I recommend, because everything is inside one place. You've got custom workflows for all of the work that you want to do day-to-day. I mean, for example, for me, again, like the best things to build with DeepSeek Harness are the things that you do every single day. So don't go after hypothetical jobs.

Don't try and automate something you don't need. Focus on what do you do every day, and then how can you automate that? So for me, for example, I was spending too much time doing keyword research. So we created this keyword research system.

I was spending, for example, too much time creating content. So we created this system for deploying and creating content. That's what the game is about, right? It's understanding where you're spending your time, and then how can you automate it so you never have to do it again?

And that's what you can do with DeepSeek Harness. And then we have our memory system over here. And so what you'll see is, like, it gets updated as we use it. So 28 minutes ago, one hour ago, six hours ago, right?

It's not updating on every single task, but for the useful stuff that it's actually spending time on that we need for later, it will jot it down into the memory. And then the cool thing about this is we have the graph where we can plug this into any harness, but then also we have the agents updating and adding to the memory as well. So it's a positive feedback loop. So that's basically it for the agentic operating system, how it works, why you'd use it, and how powerful it is.

So if we come back to that mission control example where we're building out mission control, you can see here that it's opened up and we have the full example of what we created. So if we come back, for example, just to recap, we've said, okay, create a beautiful mission control for this, then we said, okay, go off and do this. And we selected the options. It went off, just ran autonomously.

I actually went out to the cinema and just watched a movie. And then I came back and it's been fully completed. So it is really autonomous. You can see it actually checks for a to-do list, which I'll explain in a second.

And then you can see this is ready to go. Now in between that, basically what I actually did is it gave me the full details here. And if you're like, oh, I'm not sure, cause you know, it's quite a long answer. Like, I mean, you're looking at like a thousand different characters right there, right?

So once it's completed, you can say like open up and then it will actually open up and control this in your browser. So you can see an example right here of our sessions and memory, our command, right? And so we've built out a full command deck using the system from DeepSeek. So if we have a look here, we can see all our previous sessions for DeepSeek, which is pretty wild when you think about it.

It's just build that into a beautiful system. And it's actually organized my Obsidian Vault, as you can see right here. So we've got like information about me. We have the wins.

We have, for example, decision logs. These are actually, this is actually taken from my Obsidian Vault. We can even search the vault and it's done this fully autonomously without me doing anything. So one thing I will tell you is like, this is how much of a difference the harness actually makes.

When I've tried to do this myself, like, you know, if I'm using Claude or something like that, it will come back with something very, very basic. But if you look at that system right here, so we've pulled this up side by side, this is the mission control built. And then over on the left-hand side with the harness. Now he's actually built something really complex and quite interesting.

So if we open this up, it shows us how many tokens out we've built, how many sessions we've run, how many workspaces we have, how many tokens in the peak context. The model we're actually using with DeepSea Harness because this is a mission control. And so we've basically taken the basic harness over here and turned it into a beautiful mission control. Then we can see our recent sessions.

So if we look through here, we can see our recent sessions. We have our workspaces. We have our previous session history. We have our memory and we have the command engine where we can actually just run headless runs with DeepSea.

So what that means essentially is that we can run a command and then we'll see the output over here. So if we say, for example, just a terrible example, but let me give an example, like build out a snake game. So we'll just, you know, we'll vibe code like it was 2024 and then we'll hit launch like so. And that's actually going to launch a session right here with DeepSea Harness.

So we've basically put like a interesting mission control system on top of the DeepSea Harness in like one single prompt. So it's done a very, very good job, which actually blows me away by how interesting it is and that sort of thing. Now, another thing that we were talking about previously was, for example, models. So if we go inside the edit section here, we can plug in an API key, but we can also add a model here.

So you can actually go inside the creator mode like so. start a new session like this, creator mode and then we can say okay can you add OpenCode as a drop-down with DeepSea V4 Flash that I can use the harness for free when I am building with it. So you see how we're using the power of creator mode to start customizing the plugins and the way this is built and so we can actually rewrite our workspace and customize it in this way and then just hit enter over here and that should build out a plugin where we can have a drop-down box to use OpenCode locally, well not locally but to use our local installation of OpenCode and then to plug in that free model where we can use DeepSea V4 Flash for free and so you got that build on the left hand side and over here we have the Snake Game running on the right hand side so you can see that's been fully built over here okay so that's been created and then we have our commands so we can do help, workspaces, session status, version and memory, pretty nice. So it is really amazing, really interesting as a project and you can see how that's building out right here also something else that you can do here is you can have multiple sessions coding at the same time so if you were looking at this mission control system here what you could actually do is you could have multiple agents multiple conversations running on the same process so if we post a iteration or you know we post a prompt here we'll actually have two different agents two different conversations inside the same project the same for example with Obsidian we can have two conversations running at the same time and they can just go side by side and you can have like multiple of those you could have like six different agent conversations coding into the same folder to build out like multiple features at once so to give you an example this right we could do new session here inside the same folder so we're running inside the mission control and if you want to change the folder you're working in you can just click the drop down select between your recently used or click on add a workspace add a new one like so and then over here we'll say okay make the mission control like really bright and colorful and interesting so what we can do here is we've now got the deep seek section that's being built in with open code and over here we have the mission control being repackaged in terms of the ui and the colors so you see how we've got multiple conversations going on at the same time to really improve the systems and make them even better so you can do that inside the same project or you can do that outside the same project as well which is really interesting so they're basically coding out in the background right here and the other cool thing about having the agentic operating system is that you can open it in full screen right here right so you see how we've got the link at the top this is where it's actually running on a local host but we can use it embedded inside our operating system as well which is great one thing that i also like to do is i like to switch between two different styles so i have deep sea coder over here which is just using open code with deep sea fee for flash so we can code for free and then we also have the harness which is free but you would use you know different models inside there so those being built out right and then as that gets updated you can just refresh this and it will begin to be built out directly right there which is quite nice now if we have a look at the harness here we've pretty much covered everything when it comes to this i mean there's little things like you can close and open the sidebar you can add new sessions you've got the set settings here and you can choose how you group it by whether you want it manual or last updated and that sort of thing you can also delete a project here so we can click on delete workspace and just remove that if we want to and we have the ungrouped chats over here too as you can see and sometimes what you see here is like it gives you a local address like this you can either open that up manually and just copy and paste that into your browser like that or the other option you have is you can just say you know open this up or turn this into a web page so i can view it view it and preview it and that sort of thing which is quite nice so these are the finishing touches and then you can see over here it's asking us for approval so we're going to click on allow and that's beginning to edit the settings inside the dot dsh folder which is where the visual ui of how it's being written is is right there right so the dot dsh if you remember earlier is the folder where the ui and everything that we build inside creator mode is is published and you can see here for example it's editing the public forward slash style css and i think it's quite interesting as well because it sort of teaches you what it's doing and how it's coding but you don't need to be technical to use it so you can you can just keep it simple and give it simple like plain text prompts but also you kind of learn like what it's doing and how it's cutting out the project and where it's putting everything and why it's doing that and that sort of thing and also if you click on these you can see it's thinking so you can see it's reasoning as it goes through right so it's like here's the css now let me consider this actually let me think about this so if you want to go like really in depth in terms of what it's doing you can view over there and now we can see we've got the new version of this setup as you can see right this you can switch between um all these different settings and it's just improved it slightly i actually don't like the way that it's redone that but yeah you get the point like you can just iterate and improve and stuff like that one other thing that you can do here is you can build out skills so if we go to the settings we can actually teach skills to our ai agents like a harness so what does that mean so basically like skills are also plugins inside the setup that we've got here okay so when we're for example like running the seo workflow we can say okay save this is a plugin do that as a plugin right so for example here let's say i've got a skill for seo or let's say for example we have a ui skill like ui ux pro max we can grab that then we can go over to the agentic os and say can you install this as a plugin as a skill so that you can actually create really nice ui so if you see for example we can actually teach it as a skill and we can improve that directly there as well also pretty cool so you know i was talking before about how to add deepseek v4 flash for free with open code so if we click on this drop down now and we click on the model you can see that we've got deepseek v4 flash deepseek v4 pro and now we have added deepseek v4 flash free how cool is that so we've added a free model inside a harness and that's just ready to go right there so if we actually look at what it's done over here we said okay can you add open code as a drop down with deepseek v4 flash so that i can use a harness for free when i'm building it said sure i'll investigate that now let me build it and then it says the key harness here's the information and then it started looking at what open code is so it was like reading and thinking and have a look through our files they said i've got the new the full pitch here right so your harness ships the open code provider but it's mounted dormant so it only appears in the model drop down once the setting profile activates it so open code zen offers deepseek v4 flash for free and so it needs an open code setup and then it's rooted that inside the settings so that it's a model that we can add on the drop down all right so it says it ships with the drop down here and then we've got that plugged into the system pretty cool actually amazing that you can just change the drop down like that i mean you could do the same for example if you wanted to code for free you could also go to open router you could go to omni router you can actually type in free on open router here and you'll see all these different models that you can then plug in directly into deepseek harness to code for free as well so that's basically it that is the full course we've talked about all the different levels the top levels agentic operating systems the basic level is just using like the coding functionality to build out whatever you want we've covered skills plugins cordis why it's so different what makes it useful versus everyone else and also how to to get the most out of it and even add your own free models and how to plug it into an agentic operating system where you've got everything inside one place now if you want to get more training on this sort of stuff if you want to connect with me personally feel free to join the ai profit boarding link in the comments description or just go to the aiprofitbomb.com you see we've got a bunch of nice reviews and testimonials right there inside the community you can post questions get help and support in real time and i personally answer those questions on a daily basis in the classroom you can actually get access to all of our best trainings so we add new courses if you like these sort of courses we've got a bunch of courses over here we have all my playbooks i personally use inside my own business and then also inside the new daily update section you can grab the agentic operating system that we've built and we've just updated it plus you can get the zip file to install it and then you can get access to all of our new trainings and useful materials right here inside the calendar you can jump a weekly coaching course so there are four weekly coaching course per week where you can ask questions and get help and support in real time and in the map you can meet people in your local area who are building out with ai agents like you so see inside there link in the comments description or just go to the aipuffbomb.com that is the full deep seek harness course as promised and excited to see what you build cheers for watching