0s This is the full Deep Sea Carnis course
3s and by the end of this video, you'll
5s have a free AI employee working at your
8s machine 24/7. Deepc just dropped a brand
10s new tool. It's completely free and the
12s thing turns your AI from a chatbot into
14s a real worker that reads your files,
16s writes your drafts, and does your actual
17s jobs for you whilst you're at the gym.
19s It's one of the best coding harnesses
21s out there. One of the fastest growing
22s GitHub projects ever. And I want to take
24s you through the three levels to learn
27s this bad boy. So level one, you'll have
29s it running in under 5 minutes. Level
31s two, I'll show you how to build your own
32s custom tools and automations, apps,
35s websites, just by typing one single
37s sentence. And there's a hidden setting
39s inside this level that almost nobody
41s will ever find, which I'll reveal later.
43s Level number three is the endgame. Your
45s own team of AI agents all working
47s together whilst you sleep with Deep Seek
49s Harness. So stick with me until the end
51s and I promise you'll walk away with your
52s own free AI employee, no matter who you
55s are. Let's get into it. Now, quick facts
57s before we start so you know what we're
59s dealing with here. Deep Seek Carnis came
60s out on August 13th as a developer
63s preview. It's completely free open
65s source MIT license which means you can
67s download it, you can change it, you can
68s even build a product on top of it
70s without asking for anyone's permission
72s and also it blew past 115,000
77s stars on GitHub which puts it among the
79s fastest growing AI projects ever
81s recorded. There already over 300
83s community plugins for it sorted into
85s categories like knowledge, media,
88s workflows, integrations on a tool that's
90s just days old. The project is less led
93s by Tani Sui who joined Deepseek in March
96s 2026 from Jane Street, one of the
98s sharpest technical firms in the world.
100s And a team of 19 people took this from a
102s sounding star in May to a public release
105s in August. Also, some people will be
107s watching this and be like, "Well, the
108s harness is free, but the brain isn't
110s free because you have to pay for the
111s API." I'll show you a way around that to
113s get a free brain plugged in later in
116s this course. Let's get into it.
119s So, let's get into level one, which is
121s installing it, understanding it, and
122s putting it to work. First of all, one
124s minute on what an harness actually is.
127s If you get this one idea, the entire
128s rest of this course clicks into place.
130s So, don't skip it, even if you think you
132s know. So, an AI model is a brain, just a
135s brain. And it can think, but it cannot
138s touch anything. It can't open your
140s files. It can't remember what you told
142s it last Tuesday. It can't use a browser
144s or run a program or save it work on its
147s own. A model is a very smart voice in
149s the jar. The harness is everything
151s wrapped around that brain. So, it's the
154s hands, the memory, the workspace, the
155s rules about what it's allowed to touch.
157s When people say AI agent, what they
159s actually mean is model plus harness,
162s brain plus body. So for example, Claude
166s code is a harness around Claude. Hermes
168s agent is a harness around any agent
170s model that you plug into it. The model
172s gets a credit, but the harness is where
174s the work actually happens. And harnesses
176s matter more than most people think. So a
178s top model might be 5 or 10% better than
181s the one below it. A good harness makes
184s the same model dramatically more useful
186s because it's a harness that decides what
188s the model can see, what it can do, and
190s how it recovers when something goes
192s wrong. The industry spent 2 years
195s obsessing over brains, but now it's
197s really moved to bodies. So, Deep Seek
200s Harness is Deepseek's body, right? Uh,
203s until now, the Cirrus harnesses were
205s mostly closed products. So for example,
207s you had claude code, you had codeex and
210s you could switch the brains around those
212s in though in them, but Deep Seek Carnis
215s is really designed to be open source and
217s easy to switch around the models and
218s pretty much switch everything around it.
220s Right? So on August 13th, Deep Seek
222s published a direct alternative to that
224s whole ecosystem of Claude Code and
226s Codeex and Hermes and everything else.
228s And this is MIT license. It's on GitHub
230s for everyone. And the same lab that
232s shook the model world in early 2025 just
235s did the same thing to the harness world
237s this year. And that's a context. Now
239s let's actually get it on your machine.
241s So step number one, no.js. This is the
243s only thing you need installed first.
245s No.JS is a free program that lets this
247s kind of software run on your computer.
249s You can go to no.js.org. Download the
251s version for your system and click
253s through the installer like any normal
254s program. Next. Next. Finish. 2 minutes.
257s If you already have it, skip ahead. Step
259s number two is you just want to open this
261s up inside your terminal. So on a Mac you
263s can press command plus space type
265s terminal hit enter. On Windows check
268s start, type PowerShell, hit enter. I
270s know the terminal not the nicest place.
272s I know it's like a bit intimidating for
274s some people and non-technical but you're
276s just going to type one line and after
277s today you may never think about the
278s terminal again because this tool doesn't
280s actually live there. And step number
282s three is you just type this command and
284s hit enter to install it. That's it.
287s literally this command hit enter inside
289s your terminal just like you can see
291s right here and you are good to go my
293s friends and once you've set it up it's
295s going to look something like this and
296s that's a whole install so that command
298s downloads DC caris and starts it the
300s first run takes a minute or two whilst
301s it pulls everything down and when it's
304s ready it gives you a local address by
306s default is your own machine port which
307s is 3080 and it opens the page in your
310s normal web browser I actually like this
312s about DeepSeek I think it's quite a
314s smart choice which is that it doesn't
316s keep you in the terminal, right? It runs
318s as like almost like a normal web page.
320s So, you have a chat window, a sidebar,
321s buttons, a folder picker. It looks like
323s software you already know how to use.
325s So, you can see ours running over here.
327s Almost every other Sirius harness lives
330s inside the black terminal window, which
332s quietly filters out everyone who isn't
333s an engineer. You can, for example, use
335s apps like Claude, but again, like it's
337s another app. It's another button you
339s have to click outside of your ecosystem
341s because most people are just directly
342s inside the browser for most of the work
344s they do. With this system, it's just
346s like inside one page and you can just
348s run it and have it ready to go or you
350s can plug it into an agent operating
351s system like you see right here and it's
353s way easier to use. I think DeepS shipped
356s this web first on both of us, right?
358s Because the page runs locally on your
359s machine, but it's not a website someone
361s else hosts, it's yours. And then step
363s number four is you want to connect a
365s brain. So the first time the harness
367s opens, as you can see right here, it
369s asks for a Deepseek API key. An API key
371s is just a password that connects the
373s harness to your model. You can go to
375s platform.deseek.com,
377s sign up, click API keys, create one,
378s copy and paste it into the harness and
381s that's done. That connects you to Deep
382s See models including V4 Pro, their
384s flagship built heavily for agentic work
387s which shipped the very same day as a
388s harness. So the brain and the body
390s launched together. They were built for
392s each other. Now you can also change the
394s economics of this as well because inside
396s your settings under models there's an
398s add provider button which can connect
400s open router. So you can use a free brain
402s inside this. So for example if you
404s connect open router that uses cheap and
406s free models. If you use lama with the
408s deepc carnis you can run free local
410s models with this if you've got a decent
411s setup. You could also for example find
413s find a free model of open rout plug that
416s in or you could use deepc v4 pro uh
418s which is their flagship model and that's
420s still pretty cheap but it's not free.
421s And then you can add anything else. So
423s you can just go inside your settings
424s over here and then we can go to models
426s and then we can add a provider like so.
429s We can flick down the list and we could
431s be like okay I want to use open code.
433s Now here's what's pretty cool about open
435s code. Open code has free models and
437s guess what one of its free models is
439s deepseat v4 flash. So you can actually
441s get a free provider like open code. Plug
444s in the API key from open code and then
446s as I promised earlier that's how you can
448s use free models with it. And then the
450s harness is free and the brain is free
452s and you've got a full free system ready
454s to go. Now you can also give it a
456s workspace. So in the interface if you
458s click the folder icon you can add a
460s workspace. This is pretty useful. So let
462s me explain it properly. Basically the
464s workspace is the room the agent works
466s in. So you see how we can select a
468s folder that we want it to run in and we
470s can add a workspace like this and then
472s just select a local setup. Well
474s basically this is the room that your
476s agent works in. So it can read every
478s file in that folder. It can create files
479s in that folder. It can change the files
480s in that folder. It can't touch anything
482s outside it. So you don't want to point
484s at your whole computer because you want
486s to make one clean folder, right? So
488s don't point it at your whole desktop. I
490s have over like 10 years of clutter on
491s it. You want to make one clean folder.
493s Call it agent workspace or whatever you
494s like and put copies of the files you
497s want it working on inside. So that could
499s be like your draft content, your
500s research notes, your meeting notes,
502s whatever your actual work is. So it's
504s one room with a door you control. And
506s this keeps things safe. And also just as
508s important, it also keeps the agent
510s focused on one single place because you
513s don't want like an agent digging through
515s thousands of irrelevant files. That
517s would actually cost a lot of tokens,
519s cost you time, and also it would be
521s super slow and unfocused. So instead,
524s create the workspace, make it nice and
526s clean, one clean folder. And then step
528s number six is your first real task. And
530s you don't want to just, you know, write
532s a poem with this or something like that,
533s right? you you want a real piece of your
535s actual work because that's the only way
536s you can really start getting the most
538s out of this. So, look at what you spend
539s your time on, look at something that you
541s spend on every single day or a cool idea
542s that you have and then you can start
544s building with this directly, right? And
547s the way that you use Deep Seeker, as you
548s can see right here with the harness, is
550s you just write in plain English what you
552s want, exactly the way you'd brief a
553s smart new assistant who was working for
555s you. So, here's a real example. We could
557s say, "Read all the notes in this folder.
559s Write a one-page summary of the main
560s points as a new file called summary.MD.
563s use simple language and short bullet
565s points and then you can see for example
567s here that it comes back with all the
569s responses [snorts] and this is a harness
572s that is agentic so it can run on long
575s horizon tasks it can do stuff on your
577s computer it's not just a chat window it
580s actually thinks it lists the files in
581s the folder to see what's there and it
582s opens up one by one and it works through
584s the content and then it can actually
585s write a new file into your workspace so
587s the great thing about this is you can
589s use it for building apps for building
590s tools for building websites for building
592s for example games even right a harness
594s does work in the place the work lives
597s and the output is something real that
599s you can open send or build up so let me
601s give the pattern for writing good tasks
602s because the quality of what you get back
604s is mostly decided by the quality of what
606s you type in so there are three parts to
608s this what to look at so for example like
610s read the notes in this folder what to
612s produce so for example a onepage summary
614s as a new file called summary MD and what
616s rules to follow so for example simple
618s language or short bullet points right
621s look at produce rules Uh that's a whole
623s skill and [snorts] you don't need like
625s magical prompts. You just need clear
627s briefs like you just talk to this the
629s way you would message a friend. So if we
630s wanted to create a new session here for
632s example, we can just click on new
633s session and we say okay build me a
635s beautiful website for my SEO agency and
637s use my local Obsidian folder to come up
640s with context so that you understand what
642s my SEO agency is about. And then we can
643s plug that in. And then we can select the
645s folder we want it to work from. So for
647s example here we could go inside uh
650s Obsidian Vault. We can plug that in. We
652s can click choose. And then we can hit
654s enter like so. And that's going to begin
656s to work its magic. And that's the whole
657s skill. So you don't need anything fancy
659s here. If your instructions would confuse
662s a human assistant though, they'll
663s confuse the agent. So you want to make
665s this clear, easy to understand, give you
668s access to relevant local files, and just
670s make it very easy to use. And you can
672s use this for like content creation. You
673s can see how fast it runs as well. So it
675s runs really quickly. We can also see a
677s full session log over here. We can add
679s new sessions. So we can see multiple
681s projects that we're working from, what's
683s ungrouped as well. And then we have like
685s our plugins over here too. So plugins
688s for example, we've got like web search
690s and we can allow it to search a web. We
692s can set up agent loops. So how many
694s parallel tool calls it runs. We have the
696s shell so shell command timeout as well
699s and plug-in list. So we can actually add
701s new plugins that can access tools from
704s outside. So for example over here we
706s have like web enabled which means that I
709s can actually search web using a plugin
711s that we've installed over here which is
712s pretty nice. We'll come on to more stuff
713s like that and like you know more complex
715s things on that as you can see. Now also
717s you'll see as we use it it's like who's
719s this site's primary audience. We can
720s click on this for example we can say
722s what's the single main action that we
723s want uh book strategy session how do we
725s want it to look so modern minimal SAS
727s and then also one landing page. So you
730s can see that it's going to give you
731s options to select. So, where should I
733s build the site files? For example, we
735s can select document folder. Hit submit.
737s And now it's deep diving through
740s everything that we're creating locally
742s here. Pretty nice. Now, at this point,
744s you might be wondering, okay, you know,
745s where do you start with this? How do you
747s get the most value out of here, you
749s know, all these tools, all these
751s different apps. I know a lot of people
753s who are not focused and not getting the
756s n out of these systems. And that's what
758s I don't recommend to you. So the main
760s thing here is that you give this harness
763s real jobs from day one. And if you're
766s not sure what real jobs are, look at
768s what you're working on and then go from
770s there. So for example, if you make
772s content, drop your notes in the
774s workspace and help it create content for
777s you. You know, that could be, for
778s example, if you do SEO, get it to
780s automate your SEO for you. Whatever you
782s work on dayto-day, that's what you
784s should be giving to Deep Sea Carnis.
785s Don't use it to just vibe code a snake
787s game or something. So the main point
789s here is like you make it actionable. You
791s make it actually useful and you can just
793s focus on one automation per week. So for
795s example, for me, what I like to do is
797s I'll measure my time. I'll measure what
799s I work on dayto-day and I'll put that as
801s a list and then I'll be like, right,
803s okay, well out of all these tasks, this
805s will be the easiest for me to automate
807s next with something like Deep Sea
808s Carnis. So, if I'm looking at this, I'm
810s like, right, okay, well, each day I
812s create content for my website. So, let's
815s make sure we automate on a schedule
817s creating content on my website using
819s Deep Sea Caris instead of me creating it
821s daily. And that [snorts] way, we
823s actually turn it into something useful
825s instead of just going after like random
827s use cases that don't make any sense.
828s Now, let's also talk about a useful
831s feature that's quite unique to Deepseek,
834s and it's called the trajectory view.
836s Right? the trajectory view as you can
839s see right here. And the way that this
841s works is quite interesting. So next to
844s your session there's a trajectory panel.
846s You can click it and what you get is
848s every single step the agent took laid
852s out panel graphically clickable in
856s order. So if we have a look for example
857s here we can take a look at the
859s trajectory. So these are two different
861s versions. The chat is just plain text.
864s You just talk to it in plain text. The
866s trajectory is you can see what's going
868s on and you can see it's clickable in
870s order. So for example, we can click on
872s this and we can see what's happening and
874s we've got the summary, the preview and
875s everything else. Now also as you're
877s building with this, you're going to see
879s that it asks you to reject or approve or
882s allow stuff. So as we're going along
884s like we can allow it if we've set up the
886s permissions. If you don't want to have
887s to click through that every single time,
889s you can click on full access and then
891s enable full access. Only do that if
893s you're comfortable. If you're not
894s comfortable with it, don't do it. Okay,
895s makes sense. Now, as we go through the
897s trajectory,
899s basically this works in order. So, what
901s you can see here is like the system
903s prompt it was actually given. So, if we
905s scroll right to the top inside this
907s trajectory view, we can see the initial
909s system prompt and basically this is what
912s it was given at the start. So, this is
913s your message. the context it loaded, its
915s actual thinking, the reasoning visible
917s in full, every tool it called, the exact
919s command run it ran, and the exact
922s payload it sent, the exact results it
924s came back, and how long each step took.
927s Also, it shows you, for example, how
928s many turns the whole job needed. So, you
930s can click any single step and read the
934s raw detail underneath it. Nothing is
935s like hidden. So, why does this matter so
938s much in practice beyond it like being
940s impressive and useful and looking super
942s cool? So basically AI agents make
944s mistakes all of them. Anyone who tells
946s you otherwise is basically you know they
949s haven't tried AI agents. They break all
951s the time. Agents break all the time.
953s They are easily misled, easily
956s distracted. They're not at the point
957s where they can just build and automate
959s anything without having an understanding
961s of how it works. So the difference
963s between agents you can trust with real
965s work and agents you can't is whether you
967s can see why a mistake has happened. So
969s with most tools, a failed job gives you
972s nothing. The output is wrong. Good luck.
975s Understanding why. With the trajectory
978s view, you can scroll to the exact step
980s where things went sideways. So you can
983s look through here and you'll be like,
984s "Ah, it looked in the wrong folder. Ah,
986s I misread that one line inside my brief.
989s Ah, the file it needed was empty." So
990s you can fix that one specific thing and
993s then run it again. So we can have a
995s look, for example, here we can see what
997s it did, what it worked on. Um, and then
999s if we want to, we can basically take
1001s that and then run it again based on
1004s where it's broken. This is also the
1006s complete opposite of the direction that
1007s a lot of closed labs have taken where
1010s basically the model's thinking. It's
1011s kind of like hidden from you. You know,
1013s you can see what's going on, but you
1014s will never get as much detail as this.
1016s Like this is super super useful for um
1019s compared to a lot of steps. So Deepseek
1021s made every step visible. It's kind of
1023s like a care core design like principle.
1025s So once you've managed an agent, you get
1027s the full vision. You can see everything
1029s that's going on. Um and it basically
1032s gives you a lot of visibility and
1033s understanding. Now also what you can do
1035s and this is another level is you can get
1038s the session log. So you can actually
1040s export the full session as a zip file
1042s and inside is a complete log of every
1045s step. So you can export any session as a
1048s complete log. And with this you can see
1051s every step, every call, every result in
1053s a standard file format. So you can save
1055s it and you have a permanent record of
1057s exactly what the agent did. If you later
1059s hand agent work to a virtual assistant
1061s or a team member, that log can help you
1063s check the work without having to redo
1065s the work. So you review the trajectory
1068s the way a manager reviews a report. Last
1071s thing in level one, by the way, and this
1073s is still level one here, is this version
1076s is 0.1. And what I mean by that is
1078s basically Deepseek have only just
1080s released this. This is 0.1. Basically,
1082s it's 10% of the full version. And so,
1084s it's a developer preview. And what this
1086s means essentially is like be expected
1089s for things to be a little bit bump.
1090s Anything open source doesn't always work
1092s first time. So, like it's pretty cool
1094s that we can see the chat, we can see the
1095s trajectory. It can build awesome stuff
1097s as we've already seen. It's pretty easy
1099s to use, but bear in mind like there's
1101s going to be things that break. There's
1103s going to be things that move around in
1105s between versions and updates. A plugin
1107s that works today might break next month
1109s and get fixed a month after. So, if you
1111s wire your entire operation to it this
1113s week and treat it as finished software,
1115s something's going to break. And I just
1117s want you I want to be honest with you,
1118s right? Uh because you can start doing
1120s real jobs with it, but basically, you
1123s want to structure it around a system.
1124s And I'll show you that later inside
1126s level three. And it's one of the
1129s strongest ways to use a So, you want to
1132s build this into a system, which we'll
1133s come on to later. But that's essentially
1136s level one complete. Well done for making
1138s it this far. So one key connects to
1140s brain, one folder is the room it works
1143s in. Playing English briefs with a look
1146s at uh produce rules system full
1149s trajectory vision on every step session
1152s logs so you can keep and check
1155s everything and basically if you stopped
1157s right now you would have a free working
1160s fully inspectable AI agent doing real
1161s tasks on your machine today but you'd
1163s only be using 20% of the full power of
1166s deep sea. The other 80% is the part that
1169s made me make this course. So that's
1171s level one. Now quick pause here before
1173s we get into the rest of it. Everything
1175s in this course we run live inside the
1177s air puffer boardroom and the deepseat
1179s harness is on the daily tutorial
1182s schedule right now. The exact install
1183s from level one connecting free models
1185s for open routters. The agent work cost
1187s nothing. Um and also just like how to
1189s get the most out of a automation. So if
1190s you want really practical useful stuff
1193s like this, if you want weekly coaching
1194s calls, if you want to connect with
1195s amazing people, feel free to check out
1197s the AI profit boardroom link in the
1199s comments description or go to the air
1201s profitbomb.com. Every single day I
1203s answer your questions. I create
1204s personalized video tutorials for you. I
1207s have a classroom with all this amazing
1209s courses. You can jump a weekly coaching
1211s course right here. You can meet people
1212s in your local area who are building out
1214s with AI agents and stuff like this. And
1216s it's just a great place to learn about
1218s DeepS AI agents. You can meet cool
1221s people and it's a fantastic community.
1223s So with that, let's get into level two.
1226s Level two, my friends. The plug-in
1228s system, the four mo modes, creator mode,
1231s the hidden config file, how the model
1234s works, and the one line in the docs that
1236s quietly connects this tool to every
1237s other agent that you own. So let's start
1239s with the headline of DeepC's own GitHub,
1242s which is three words. Everything is a
1244s plugin. Okay, so most software companies
1247s say something like that and mean like we
1249s have an app store where you can add
1251s extras. Deepseek means it literally and
1253s the difference is the whole story of
1255s this tool. So the sidebar and the
1256s interface is a plugin. The tool the
1258s agent uses to act the plugins the system
1260s prompt the standin instructions the
1262s agent starts with a plugin. The sandbox
1265s where it works a plugin. Even the agent
1268s loop itself the core cycle think act
1271s check repeat is a plugin. So, every
1273s single organ of this machine can be
1275s switched off, swapped out, or replaced
1277s with a better one without breaking the
1279s rest. And here's the picture to hold in
1281s your mind, my friends. So, every other
1284s agent product is like a sealed car. You
1287s can change your radio station. You can
1289s maybe add like a phone mount to it, but
1292s the engine is the engine. The seats are
1293s the seats. And if the company that built
1295s it makes a bad choice, you're stuck with
1296s it. That's it. But Deep Sea Carnish is a
1299s car where the engine, the wheels, the
1302s seats, and the steering wheel all pop
1303s out and click back in and the car keeps
1305s driving whilst you do it. So when
1306s someone else builds like a better memory
1308s system next month, you swap that one
1310s part in. When a better browsing tool
1312s comes out and gets shipped, same thing.
1314s So you don't wait for the company, you
1316s don't migrate to a new product, you
1318s upgrade one piece and keep moving and
1320s someone will build better parts, right?
1322s That's what the community is all about.
1324s So let me give you some examples within
1326s hours of this launching. And again, it
1327s only launched like this week. Deep sea
1329s carnis. It's already happening at a pace
1331s I've never seen before. Right. So, for
1333s example, within hours, the community
1335s shipped a desktop version and a curated
1337s plug-in list. There are over 300
1339s community plugins indexed right now from
1341s the official plug-in tag sorted into
1343s categories. You got knowledge, media,
1345s workflow, integrations. Deepsee even
1347s tells plug-in builders to tag their
1349s repositories with DSH plug-in so
1353s everyone can find them. And when a free
1355s tool grows an ecosystem in its first
1358s days, that really tells you where the
1359s builders believe the future is, right? A
1360s lot of developers are switching over to
1362s Deep Seek Harness because it's got that
1364s open source vibe. It's so easy to share
1366s everything. So, under the hood, this
1368s whole design runs on something called
1370s Cordis and Deepseek did something labs
1373s almost never do for a product like this.
1374s They actually published an actual
1375s research paper explaining the design.
1377s Now, if you haven't seen that already,
1379s you can actually find it on the GitHub
1380s page, right? So they have a full like
1382s paper on cordis and how it works as a
1384s paradigm and essentially you don't need
1387s to read it. It is kind of interesting
1389s but it's very very heavy maths but the
1391s core idea matters and it's pretty simple
1393s to say right. So every action in the
1395s system can be cleanly reversed if you
1398s add a piece and there's a guaranteed
1399s clean way to remove that piece with no
1401s leftover mess and no side effects and
1403s everything else. Well, that's the reason
1405s everything is a plug-in actually works
1406s here, right? So instead of collapsing
1408s into chaos, which is what usually
1410s happens when software gets it modular,
1412s the swapping is pretty safe by design
1414s mathematically not held together with
1416s like tape and hope. Uh there's also very
1419s Jane Street. So if you know that world,
1422s building the boring and rigorous
1423s foundations first so the exciting stuff
1425s on top doesn't fall over. That's what
1427s it's all about. It's just about building
1429s good foundations inside this harness.
1430s And again, like I said, it's very
1432s unique. Now, you might be wondering,
1433s okay, what are the four modes? Because
1435s this is the control panel. people click
1437s past on day one and never revisit. So if
1439s we have a look for example here, you can
1441s see here that we've got this standard
1442s mode right here at the top. Right? So we
1446s can switch through this when we start a
1448s new session. So we've got standard mode,
1449s code mode, minimal mode, and creator
1451s mode. [snorts] So in settings, there's
1453s an agent preset option with four choices
1455s and each one is just a really different
1457s amount of power handed to the agent.
1459s Standard mode is home base, balanced
1462s capabilities, good judgment about tools,
1464s right? For everyday business tasks, the
1466s summaries, the drafts, the research jobs
1468s from level one. You'll spend most of
1470s your time here to be honest. Code mode,
1472s right? That actually unlocks the heavier
1475s building powers for bigger technical
1476s jobs. So, if you're having it construct
1479s something complex over many steps, this
1482s gives it more room to work. Most
1484s non-technical users will rarely use it,
1488s and that's fine. Knowing you don't need
1490s a mode is probably as useful as knowing
1492s you do. And then for example, you have
1496s uh minimal mode. And what this does is
1498s actually strips the agent down to almost
1500s nothing. It's very light. It's very
1502s fast. Fewer moving parts. Perfect for
1504s quick, simple questions where you don't
1507s need it opening files or running tools
1509s for you. Think of it as a difference
1511s between calling a whole team into a
1513s meeting and just asking the person at
1516s the next desk. And then there's creator
1518s mode. So, creative mode exists for
1519s exactly one job, and it's a feature I
1520s teased at the top of this course. So,
1523s building new plugins by prompting. You
1525s can describe a tool that you wish
1527s existed in one plain English sentence,
1529s and the harness builds that tool live
1531s inside your session and actually
1532s switches it on. So, let me walk you
1534s through this slowly because the first
1536s time you see it, it doesn't really feel
1537s real, but if you start a new session,
1539s you can pick creator mode. So, let's try
1542s this out. So, we're going to click on
1544s create a mode over here, like so. We'll
1546s select that. Make sure you click on a
1548s new session. So you click the plus sign
1550s to start a new session. And then you're
1551s going to select create a mode. And from
1554s there, you type a sentence describing
1557s something that you want to exist. So
1560s let's make it concrete. In this example,
1562s we actually said, add a small panel in
1565s the bottom corner that shows my three
1566s tasks for today and let me tick them
1568s off. So you can now send that as you can
1572s see right here. And then we can click on
1574s full access and we'll just allow that as
1577s you can see right here. And basically
1579s the moment you send that the harness
1580s automatically loads its own plug-in
1582s development skills. So you can watch it
1583s do this in the trajectory view. So if we
1585s go to the trajectory view that we were
1586s talking about before, you can see here
1588s that it's beginning to use this system
1591s right here. Right? And so if you look at
1593s this example, this is a sped up version.
1596s Basically when we've asked for the
1598s system, it's got the daily panel over
1601s here. Right. So, we asked it to add a
1604s small panel. This is again sped up just
1606s so you can see it in in action. But the
1608s main thing here is like you can build
1610s your own workspace. Like how amazing is
1612s that? You imagine you can go into Claude
1615s and you're like, "Mate, I want this to
1616s look like that. I want this to look like
1618s that. I want to build my mission
1620s control." You can't do that inside
1621s Claude. It's closed. It's not editable.
1623s It doesn't swap in and out. But Cordis
1625s is unbelievable for doing this. Like,
1627s I've never seen any other tool that can
1630s do that. And I don't want to hype it up
1631s or like tell you how amazing it is
1633s without you trying it and then you try
1634s it and maybe it's not as great as it
1635s sounds, but it is impressive. Like I've
1638s never seen that before. So, what this
1639s allows you to do is basically build
1642s something in real time inside the
1643s plug-in workspace and just add new cool
1646s stuff. So, if you're building out and
1647s you're like, do you know what? I wish I
1648s had a goal section. I wish I had a
1650s journal section. I wish we could do
1651s this. I wish we could do that. We can,
1652s you can do it, right? It's kind of like
1653s a genie. So, [snorts]
1655s it starts building. A minute or so
1657s later, it presents the finished plugin
1658s and asks you to approve it. One button,
1660s you click approve and the panel appears
1661s on your screen. So, you can now takeick
1663s off tasks. You didn't write a line of
1665s code here, by the way. You didn't
1667s download anything. You didn't wait for a
1668s developer. You just described what you
1670s want and the thing came into existence
1673s in front of you. How amazing is that?
1676s Now, if you don't like it, you can edit
1678s that as well. So, you can say, "Okay,
1680s make it smaller or move it left or I
1682s want different colors or keep it
1684s forever." Right? So you literally
1686s describe what you want. And so it's a
1688s very editable and sort of flexible
1691s harness that again I've never really
1693s seen before. So it's like a
1695s conversation. You can go back and forth
1696s and change it exactly how you want and
1698s mold it in the vision that you have for
1700s it. Now you can also stop the plugin and
1703s then it can basically end instantly. So
1706s you can actually switch off anything
1708s that [snorts] you have as well and you
1709s can start it and it's back. So by
1711s default it just lives inside that
1713s session. But you can also persist it as
1715s well. So it's permanently part of your
1716s harness. So you see how it's now
1717s defining the plugin. Again, this is slow
1720s motion mode versus the demo that I
1721s showed you a second ago. But that's
1723s basically how it works. And then if you
1724s build something, you can keep it
1726s forever. So you see how it's just built
1727s out this task panel now and we can
1729s approve it. And boom, look at that.
1731s We've just built out a list of goals
1734s that we want. And then if we want, we
1736s can make that persistent forever or we
1738s can just delete it. And we can add
1739s tasks. We can edit it. It will take it
1741s off and that is ready to go.
1742s Unbelievable. Really cool. I just love
1744s how flexible that is. You're never going
1746s to get that with a closed model, but
1747s this is the power of open source. It's
1749s just it's designed for what you want.
1751s Also, when you think about this, it's
1753s like, okay, well, who are they
1754s targeting? I think they're targeting
1756s people, particularly developers, who use
1758s stuff like Claude Code dayto-day, but
1760s they're like, I want it to be more my
1762s style. I want it to be customized to me.
1764s And so, this is this is something that
1766s you can do. Anyway, you can basically
1768s build stuff, keep it forever. And if you
1771s think about what this actually changes
1772s like, every piece of software you have
1774s ever used was somebody else's guess
1775s about what you need. But the menus, the
1777s buttons, the panels, it was all designed
1779s for an imaginary average user who isn't
1781s you. Creator mode flips that
1783s relationship. So the interface becomes
1784s clay. You want a panel that tracks your
1787s content published from idea to publish.
1790s Describe it. You want a button that
1792s kicks off your morning routine with one
1794s click, describe it. Right? a little
1796s dashboard showing the last five files
1798s your agents produce. You just describe
1800s it. So the gap between I wish this tool
1802s did X and the tool doing X collapses to
1805s one sentence and one approval click. And
1807s I've been doing this for a while and I
1809s still find that slightly hard to believe
1810s every time it works. Now the buried
1813s setting setting that I promise you at
1815s the start. This one's for when you want
1817s control even creator mode doesn't give
1820s you the power to switch off pieces of
1822s the product yourself. So on your machine
1824s there's a dsh folder. All right. So dsh
1828s like so. And inside it is profiles then
1831s web. And in there sits a file called
1833s cordis.patch.yml.
1836s You can open it up in any text editor.
1838s Notepad is fine. It's a short simple
1840s list. Now here's what you can do with
1842s it. You can actually write the ID of any
1844s plug-in in the system. Um remember
1846s everything is a plugin including core
1848s interface pieces like the cyber. And you
1850s can add like one indented line
1852s underneath that says disabled. True.
1854s Right. Then you save the file, reload
1856s the harness in your browser, and that
1857s piece is gone. So you can edit basically
1859s the config file inside your DSH and
1862s change it and mold it however you want
1864s or you can get the agent to do it for
1865s you. So that's what they mean by like
1867s editing the harness. You can basically
1868s change everything you want to. And then
1871s you see how we've now got the Cordis
1873s plugin section over here. And we can
1875s delete stuff. We can stop it. We can
1877s play it, run it, change it, however we
1879s want to. And it's all ready to go inside
1881s the plug-in section over there, which is
1882s pretty nice. And then if we go to
1884s plug-in list, you can see all these
1886s other plugins and the plug-in
1888s configuration, too. Now, you can also
1890s view and see how each of these presets
1893s is configured. So, if we open up this
1895s configuration file like so, we can click
1898s on that and open it up and edit it. But
1900s then we can switch between these
1901s different modes. And we can also add and
1904s draft a custom preset. So, if we go to
1907s settings here, agent presets, you've got
1909s minimal, creator, standard, and code,
1911s but you could also draft a custom preset
1914s as well if you wanted to add like a
1915s totally different mode that again is
1917s fully customizable. And that's down to
1919s your creativity. Like for me, I'm not
1920s really going to mess around with that.
1921s Obviously, it's going to help me too
1922s much. But if you wanted to add like a
1924s new mode that doesn't fit within one of
1926s these four boxes, you could easily do
1928s that as well. Now there's also these
1930s websites that are popping up with
1932s individual new plugins that you can plug
1935s into these systems as well, right? So
1937s for example, if we look there's a
1939s website called deepseecode.com/plugins
1942s and what you can actually do is browse
1944s like all the plugins that have been
1945s built inside the system. So for example,
1947s they've got like plugins for developers
1950s which is like quite technical but then
1952s you also have like these agent skills.
1954s So you see here portable agent skills
1956s for building, publishing and remixing
1958s games and that sort of thing. And then
1959s we can view it on GitHub and we can
1961s actually ask for example our agent to
1964s pull that in and install it to GitHub uh
1967s to to our agent. Now when you're
1969s installing anything from GitHub, you
1970s need to be careful. Check it yourself.
1972s Get one of your agents to check it out
1974s as well. Make sure it's legit and you're
1976s happy with it. But basically, if you
1978s wanted to install something like this,
1979s you could go over to your plug-in
1982s section inside the settings here. And
1984s then you can search for plugins and you
1986s can install new plugins. Or you can
1988s actually go inside the new section here
1989s and then just say, "Okay, install this
1991s plugin." And then you can pull in the
1993s plugins from other people's stuff. So
1995s you see how it creates almost like an
1997s app store for deep sea cares where you
2001s can share everything in the world. You
2002s know, everything is a plug-in. every
2004s single part, every single piece, you can
2006s swap it in, swap it out. So that
2008s concludes level two of this whole
2010s course. Before we actually get into
2013s level three, there's just one final bit
2015s that I actually want to mention about
2016s level two, and it's quite a line in the
2018s GitHub repo that I think matters more
2020s than the whole launch announcement. So
2022s the docs say that Deep Seek Carnis
2023s supports MCP, supports the agent client
2026s protocol, and reads agents MD and Claude
2028s MD files. So let me explain that in
2031s plain English. What that means is it's
2033s the bridge to level three because MCP is
2036s a standard plug socket for tools. It's
2039s how any agent can connect to any tool
2042s without custom wiring for each pair. So
2044s the agent client protocol is a standard
2047s way for agent software to talk to the
2050s apps around it. And agents MD and Claude
2053s MD, these are the big ones are
2055s instruction files. So little text files
2057s you leave in a folder then tell that
2060s agent about your business. So this could
2061s be your rules, your writing voice, your
2063s projects, what to always do, what to
2065s never do, where things live. Claude Code
2067s reads them, OpenClaw reads them, Hermes
2069s reads them, and now Deep Seek reads some
2071s as well. And [snorts] reads the exact
2073s same files. So if you don't have these
2075s files yet, start one today because it's
2077s the highest um potential thing you can
2079s actually do. And it shares across all
2081s your agents. Takes about 20 minutes. You
2083s can make a file called agents in your
2085s workspace folder or you can just ask
2087s your agent to set that up for you. You
2088s don't have to do it yourself. And then
2089s you can write in plain English like what
2091s your business does, who your customers
2092s are, how you like things written, short
2094s sentences, no jargon, for example,
2096s whatever your rules are, what your
2098s current projects are, and anything about
2100s the agent and what it should never do,
2102s right? And that's it. It's a briefing
2104s document, but the one you'd hand to a
2105s new hire. And then from then on, every
2107s agent that opens that folder starts
2109s every task already knowing your
2110s business. I also really like, for
2113s example, something called Obsidian.
2115s Obsidian is a fantastic way to plug in
2117s context on a more advanced level. So
2119s what this has is like a full database as
2121s you can see right here of all my notes
2122s about everything about it. So for
2124s example if I am using deepseek I can
2128s plug that in and this is more part of
2130s the agent operating system but I can
2131s plug in all my context and all my notes
2134s and memories and everything about me
2136s into one system and then that will get
2138s pulled in to our deepseek harness. And
2141s so that makes everything easier cuz then
2143s you know whatever new harness comes out
2145s tomorrow that could be even better, we
2147s can just plug this system in and make it
2148s even better. And if you look at what
2151s Deep Seek chose here, they could have
2152s had invented like their own special
2154s instruction format and built a W garden,
2156s but they did the exact opposite. So they
2158s built the harness to speak the languages
2160s the existing agent world already speaks
2162s like Claude MD, the format the biggest
2164s closed source rival created. a
2166s challenger that builds for compatibility
2168s instead of locking. It's just hoping for
2170s adaptability, right? They're hoping that
2172s developers come along, see that, and
2174s they're like, "Oh, we need that." Right?
2175s People like you watch this, and you're
2177s like, "Ah, I love the plug-in system.
2179s Let's build that in." So, this leads up
2182s to a very special question, which is, if
2184s every agent on your machine reads the
2186s same instruction files, works in the
2188s same folders, speaks the same protocols,
2190s why are you running them separately at
2192s all? And that brings us to level three,
2194s the agent operating system with the
2197s deepseek harness inside it. So let's
2199s talk first about what this problem
2201s actually solves. And I hear it from my
2203s AR profit boarding members all the time.
2205s So if you followed AI for the past year,
2207s you've probably collected tools and
2208s subscriptions. So for example, Claude
2210s Code, maybe Hermes, maybe OpenClaw, now
2212s Deep Sea Carnis. Each one arrived with a
2214s wave of excitement. Each one demanded
2216s setup time and somewhere along the way a
2218s feeling crept in. There are too many of
2220s these, too many tools, too many apps. I
2222s can't keep up. Every month there's a new
2223s one and I'm falling further behind. Some
2225s of you are feeling it about this very
2227s video. So, I want you to take that
2229s feeling apart properly because it's
2231s built on a picture that's wrong. The
2232s picture in your head might be a
2234s treadmill. Every new tool is another
2236s thing you must master. So, more tools
2238s means more falling behind. But mastery
2240s of individual tools was never the skill.
2242s The skill is running one system that new
2244s tools plug into. Um, when you have the
2247s system, a new tool isn't homework. It's
2249s a free upgrade delivered to your door.
2251s So you don't keep up with the tools,
2252s your system absorbs them. And here's
2255s what should give you real confidence in
2256s this approach, right? Deepseek have
2258s focused their entire ecosystem on this
2261s whole system, right? They they have
2263s literally built out a harness where you
2265s can swap everything in and out. And
2266s that's basically on a Gentic operating
2268s system as well. It's the same
2269s philosophy. So everything is a plug-in
2271s inside one tool. The agent OS same idea,
2273s one level up. Every tool is a plugin
2275s inside one system. and the biggest open
2276s lab in the world just validated the
2278s architecture. So, let me show you what
2280s it looks like in practice because this
2282s is exactly how I work every single day
2284s across my business. So, we've got one
2286s dashboard. I call it the mission
2288s control. And every agent sits side by
2291s side each with its own job. Uh what it's
2293s best at. So if we pull this up for
2295s example this is my gen 3 operating
2297s system and what we can do here is we can
2299s see that we've got these different
2300s workflows like the SEO workflow we've
2302s got a video agent we have the loop
2303s system we have notebook plugged in here
2305s we got a cambban board memory systems
2307s etc right everything is in one place one
2310s click away including deepseeek and
2313s deepseeek harness itself beautiful so
2316s that makes everything much easier to
2319s manage and that way we can have all of
2322s our agents and everything inside one
2324s system and I can rebuild any piece of
2326s it. And the thing that makes this a
2328s system instead of Apala apps is the
2330s shared memory with obsidian as I
2332s mentioned before. So every agent reads
2334s and writes to one obsidian bolt. If you
2336s haven't met Obsidian, it's just a folder
2339s of plain text notes on your computer.
2340s That's genuinely all it is. And that
2342s plainness is the superpower because
2344s plain text files in a folder are the one
2346s thing every single agent can read and
2348s write. No special format, no company's
2350s database. It's just like one agent
2353s learns and then every agent knows
2355s because they're all looking at the same
2356s folder. So, here's how to structure
2358s yours and how to make sure that actually
2361s works. So, I like to use uh daily
2363s projects, research, and outputs. That's
2366s the way that I organize this whole
2368s system. So, it looks beautiful like
2369s this, but it's just like organized into
2371s four steps. Daily projects, research,
2374s output, and it's just a bunch of folders
2376s with one instruction folder. The reason
2379s that I do that is because complicated
2380s vaults don't work very well. But a
2382s handful of folders, you know, one for
2384s daily notes, what happened, what's in
2385s progress, one for projects, one note per
2387s active project, and for example, what it
2390s is current status, one for research,
2392s everything the agents gather, and one
2393s for output drafts and finished work. And
2395s sitting in the top is your agents MD
2397s instructions from level two. So every
2398s agent that opens a vault gets briefed
2400s automatically. And that's the whole
2401s architecture. It's just folders and text
2404s files. Pretty simple stuff. If you can
2406s organize a filing cabinet, you can build
2408s this. So, if we come back to Deepseek
2410s and Deepseek harness here and we go back
2413s to the ideas that was pulling in before
2414s on our sessions, you can see that we've
2417s got everything pulled in for our
2419s website. So, this is the whole website
2421s that it's pulled in and it's all
2423s personalized to me, right? So, it knows
2424s my brand and everything else. Why does
2427s Deep Sea Caris have so much information
2429s about me from day one? Because it's
2431s pulling in all the context from
2432s Obsidian. So, it knows what we work on.
2434s It knows what we do. It knows who we
2436s work with. It has all of our FAQs. It
2438s understands everything about me and my
2440s business from day one because we just
2442s get that memory and we plug it into our
2444s agents. And we can do the same with
2446s Hermes agent. We can do it with all of
2447s our agents. It's it's an amazing system.
2450s Gamechanging once you understand it as
2452s well. So this is the whole system. And
2454s then for example, if I'm doing something
2456s in Claude, I can go over here. If I want
2458s to go back to Deep Seek, I can go over
2460s here. Right now, you might be wondering,
2461s okay, like how do you build out an
2463s agentic operating system like this?
2465s Well, I mean, if you want mine, you can
2466s get it inside the air profit border.
2467s It's already set up and ready to go. If
2469s you want to just build it like this, you
2472s can literally just create a new session,
2474s start a new folder. So, we can create
2476s add a workspace here. And we can say
2478s mission control like so. And then we can
2480s open that up. And then we can say, okay,
2482s create a beautiful mission control
2485s dashboard for deep carnis and also my
2488s obsidian memory locally. And then you
2490s can start with the tools you use. You
2491s know if you're using claw code day today
2493s and you want that next to Hermes you
2495s just go with the same and then from here
2497s we can select this. Now also you can
2499s switch between the models here. One
2500s thing to note so for example we can
2502s select v4 pro which is the most powerful
2505s frontier level AI the deepse has whereas
2507s for example deepc4 flash not as powerful
2510s but great for volume. So if you're like
2511s okay when do I use pro? When do I use
2513s flash? Flash is great for volume. It's
2515s pretty cheap but not that smart. Decor
2517s pro fantastic for building things. So if
2519s we're building something that's like the
2521s agentic operating system, we can just
2523s click enter like so and that can begin
2525s to build it out. Now this is basically
2527s like uh a dashboard, you know, a mission
2530s control. So it's going to code it out
2532s and then it's going to start building it
2533s locally, right? So you can see here it
2535s says mission control folder is empty, a
2538s fresh build. Let me read your canonical
2540s memory file and search for what DeepS
2542s actually is. And then it's beginning to
2544s build this out. And as you can see right
2546s here, right, it is beginning to build
2548s this in. Now you can also see what it's
2550s doing step by step. So it's got like
2551s context injection and then it's got for
2553s example thinking and then it's running
2556s terminal commands. So bash is like
2557s terminal commands and it's just running
2559s through everything step by step. We can
2560s also see what it's working on locally.
2562s So what it's doing here is actually
2564s reading through our obsidian vault
2565s understanding how our existing mission
2567s control is used to start building that
2569s as well. And then it's even for example
2571s checking out goldie bench is asking for
2573s permission locally as well on everything
2574s we're building and we're good to go on
2576s that. Right. So that might code out for
2578s like 10 15 minutes. But basically what
2581s we're getting there is a full beautiful
2583s system that we can build with the
2585s mission control. And you can do that
2586s with any agent. You don't have to do it
2587s with Deep Seek. Deepseek I'm just using
2589s cuz this course is about Deep Seek, but
2591s you could build out a mission control
2592s using uh Obsidian. You could use it with
2594s Claude or whatever you want. And then
2595s you can see here is I've mapped out the
2597s environment. So we got the mission
2598s control folder. Everything else, right?
2600s And it's like, okay, what do you want me
2601s to actually do here? So what should it
2603s be for? This is for Deep Seek Harness.
2606s And then we'll get a full control room
2608s and it'll begin to build out right as
2610s you can see right here which is pretty
2611s nice. And anytime you want to stop this
2613s we can go with it as well. You're also
2614s going to see here that as this bottom
2616s right hand corner the context used. So
2618s it show us how many tokens we use the
2620s system prompt the tools messages. So
2622s it's broken down how much context
2624s everything is taken up. So messages is
2625s actually taken up the most tokens and
2627s context right here. And then we can see
2629s okay how many steps we've used the LLMs
2632s everything else right cash here as well.
2634s So that's based on like basically
2636s reducing the amount of tokens we use by
2638s rereading the existing context we
2639s already have. And it's beginning to
2641s build this out, right? And so we can
2643s have that working in the background. We
2644s can also open up other tabs of our
2646s mission control. And I'll walk you
2647s through this. So over here we have the
2649s mission control dashboard. We have all
2651s these different open source projects
2652s like Hermes is sorry paperclipip is a
2655s system where we have basically a team
2658s like an organizational chart of all our
2659s agents working together. We have the AI
2661s Asia mastermind where we can have
2662s deepseek and all our agents speaking
2664s inside a group chat like so. We have a
2666s pipeline system here where we can go
2668s from idea to built app. Then we have for
2670s example like claude with the workspace
2672s everything is saved and we can preview
2674s it for later. We have artifacts so we
2676s can preview ultra code where we can have
2678s teams of sub aents working with claude
2680s and the anci and agent system as well.
2683s Then we have hermes here. So you see how
2685s each tab is basically like a new layer.
2687s So for DeepSeek, you can have different
2689s versions of DeepSeek and the harness or
2691s different custom workflows that you
2692s actually prefer to use and it's all
2694s slotted into one system. And if you're
2696s like this is going too fast for me or
2697s that sort of thing, don't worry. This is
2699s this is level three. But if you want
2701s like a full breakdown, we've actually
2704s got a 1hour tutorial inside the profit
2706s boardroom on exactly how we build the
2708s operating system, what each part does
2710s and how it works step by step as well.
2712s You can also see that Claude is sorry,
2714s Deepseek is pulling in the skills. So,
2716s it's got these skills here like
2718s front-end design, hallmark is for
2720s reducing the amount of tokens that
2721s actually uses. And then we've got front
2722s end design as well right here. And it's
2724s using those to to create better quality
2726s designs for our mission control
2728s dashboard. So, over here we have Hermes.
2730s We've got, you know, like the Hermes bot
2732s feature that just dropped. So, we can
2734s kind of use Hermes like WhatsApp where
2736s we have different chats. We got Hermes
2737s Apollo, which is a voice agent right
2739s here. And this just gives you a feel and
2741s a vibe for how much more fun AI can be
2743s when you build out these custom
2744s workflows, custom UIs that look
2746s beautiful inside one system. So that's
2748s what I recommend cuz everything is
2750s inside one place. You got custom
2751s workflows for all of the work that you
2753s want do dayto-day. I mean for example
2755s for me again like the best things to
2758s build with Deep Sea Caris are the things
2760s that you do every single day. So don't
2762s go after hypothetical jobs. Don't try
2764s and automate something you don't need.
2765s Focus on what do you do every day and
2767s then how can you automate that. So for
2769s me for example, I was spending too much
2770s time doing keyword research. So we
2772s created this keyword research system. I
2774s was spending for example too much time
2775s creating content. So we created this
2777s system for deploying and creating
2778s content. That's what the game is about,
2780s right? So it's it's understanding where
2781s you're spending your time and then how
2783s can you automate it so you never have to
2784s do it again. And that's what you can do
2785s with DC counters. And then we have our
2787s memory system over here. And so what
2788s you'll see is like it gets updated as we
2790s use it. So 28 minutes ago, 1 hour ago, 6
2793s hours ago, right? It's not updating on
2795s every single task, but for the useful
2797s stuff that it's actually spending time
2799s on that we need for later, it will jot
2802s it down into the the memory. And then
2804s the cool thing about this is we have the
2805s graph where we can plug this into any
2807s harness. But then also we have the
2809s agents updating and adding to the memory
2812s as well. So it's a positive feedback
2813s loop. So that's basically it for the
2815s agendic operating system, how it works,
2817s why you'd use it, and how powerful it
2818s is. So if we come back to that mission
2820s control example where we're building out
2822s mission control, you can see here that
2824s it's opened up and we have the full
2827s example of what we created. So if we
2828s come back for example, just to recap,
2830s we've said okay create a beautiful
2831s mission control for this. Then we said,
2834s okay, go off and do this and we selected
2835s the options. It went off just ran
2837s [snorts] autonomously. I actually went
2839s out to the cinema and and just watched a
2842s movie and then I came back and it's been
2844s fully completed. So it is really
2846s autonomous. So, you can see it actually
2847s checks for a to-do list, which I'll
2848s explain in a second. And then you can
2850s see this is ready to go. Now, in between
2853s that, basically, what I actually did is
2856s it gave me the full details here. And if
2858s you're like, "Oh, I'm not sure cuz you
2860s know, it's quite a long answer." Like, I
2862s mean, you're looking at like a thousand
2863s different characters right there. Right.
2865s So, once it's completed, you can say
2866s like, "Open up." And then it will
2868s actually open up and control this in
2869s your browser. So, you can see an example
2871s right here of our sessions, a memory, a
2873s command. Right? And so we've built out a
2875s full command deck using the system from
2879s Deep Seek. So if we have a look here, we
2880s can see all our previous sessions for
2883s DeepSeek, which is pretty wild when you
2884s think about it. It's just built that
2886s into a beautiful system. And it's
2887s actually organized my Obsidian vault as
2890s you can see right here. So we've got
2891s like a information about me. We have the
2894s winds. We have, for example, decision
2895s logs. These are actually this is
2897s actually taken from my Obsidian vault.
2898s We can even search the vault and it's
2900s done this fully autonomously without me
2901s doing anything. So, one thing I will
2904s tell you is like this is how much of a
2906s difference the harness actually makes.
2908s When I've tried to do this myself, like
2910s you know, if I'm using claw or something
2911s like that, it will come back with
2912s something very, very basic. But if you
2914s look at that system right here, so we
2915s pull this up side by side. This is the
2918s mission control built. And then over on
2921s the left hand side, we have the harness.
2922s Now, it's actually built something
2924s really complex and quite interesting. So
2926s if we open this up, it shows us how many
2928s tokens out we've built, how many
2929s sessions we've run, how many workspaces
2930s we have, how many tokens in the peak
2933s context, the model we're actually using
2934s with DC caris cuz this is a mission
2936s control. And so we've basically taken
2938s the basic carnis over here and turned it
2940s into a beautiful mission control. Then
2941s we can see our recent sessions. So if we
2943s look through here, we can see our recent
2944s sessions. We have our workspaces, we
2947s have our previous session history, we
2949s have our memory, and we have the command
2951s engine where we can actually just run
2953s headless runs with DC. So what that
2955s means essentially is like we can run a
2957s command and then we'll see the output
2958s over here. So if we say for example just
2960s terrible example but let me give you an
2962s example like build out a snake game. So
2964s we'll just you know we'll vibe code like
2966s it was 2024 and then we'll hit launch
2968s like so and that's actually going to
2969s launch a session right here with deep
2972s sea carnis. So we've basically put like
2974s a interesting mission control system on
2977s top of the deep sea carnis in like one
2980s single prompt. So it's done a very very
2981s good job which actually blows me away um
2984s by how interesting it is and that sort
2985s of thing. Now another thing that we were
2988s talking about previously was for example
2990s models. So if we go inside the edit
2992s section here we can plug in an API key
2995s but we can also add a model here. So you
2997s can actually go inside creator mode like
2999s so. So we can start a new session like
3001s this creator mode and then we can say
3003s okay can you add open code as a dropown
3007s with deep seed v4 flash that I can use
3008s the harness for free when I am building
3011s with it. So you see how we're using the
3013s power of creator mode to start
3017s customizing the plugins and the way this
3020s is built. And so we can actually rewrite
3023s our workspace and customize it in this
3025s way. And then just hit enter over here.
3027s And that should build out a plugin where
3030s we can have a drop-own box to use open
3033s code locally. Uh well, not locally, but
3035s to use our local installation of open
3037s code and then to plug in that free model
3039s where we can use steepc4 flash for free.
3042s So we got that building on the left hand
3043s side and over here we have the snake
3045s game running on the right hand side. So
3048s you can see that's been fully built over
3049s here. Okay. So [snorts] that's been
3051s created and then we have our commands.
3054s So we can do help, workspaces, session,
3056s status, version of memory. Pretty nice.
3058s So it is really amazing, really
3060s interesting as a project. And you can
3062s see how that's building out right here.
3063s Also something else that you can do here
3065s is you can have multiple sessions coding
3067s at the same time. So if you were looking
3069s at this mission control system here,
3071s what you could actually do is you could
3072s have multiple agents, multiple
3074s conversations running on the same
3076s process. So if we post a iteration or
3079s you know we post a prompt here we'll
3081s actually have two different agents two
3083s different conversations inside the same
3085s project. The same for example with
3086s Obsidian we can have two conversations
3088s running at the same time and they can
3089s just go side by side and you can have
3091s like multiple of those. You could have
3092s like six different uh different agent
3094s conversations coding into the same
3096s folder to build out like multiple
3098s features at once. So to give you an
3100s example this right we could do new
3102s session here inside the same folder. So,
3104s we're running inside the mission
3105s control. And if you want to change the
3107s folder that you're working in, you can
3108s just click the drop down, select between
3110s your recently used or click on add a
3112s workspace, add a new one like so. And
3114s then over here, we can say, okay, make
3115s the mission control like really bright
3117s and colorful and interesting. And so,
3119s what we can do here is we've now got the
3122s Deep Seek section that's being built in
3124s with Open Code. And over here, we have
3126s the mission control being repackaged in
3128s terms of the UI and the colors. So you
3130s see how we've got multiple conversations
3132s going on at the same time to really
3134s improve the systems and make them even
3136s better. Uh so you can do that inside the
3138s same project or you can do that outside
3140s the same project as well which is really
3142s interesting. So they're basically coding
3144s out in the background right here. And
3146s the other cool thing about having the
3147s Aentic operating system is like you can
3149s open it in full screen right here.
3152s Right. So you see how we've got the link
3154s at the top. This is where it's actually
3156s running on a local host, but we can use
3158s it embedded inside our operating system
3161s as well, which is great. One [snorts]
3163s thing that I also like to do is I like
3164s to switch between two different styles.
3166s So, I have DC coder over here, which is
3168s just using open code with DC FIFA. So,
3171s we can code for free. And then we also
3172s have the harness, which is free, but you
3174s would use, you know, different models
3176s inside there. So, those are being built
3178s out, right? And then as that gets
3179s updated, you can just refresh this and
3181s it will begin to be built out directly
3183s right there, which is quite nice. Now,
3185s if we have a look at the harness here,
3186s we've pretty much covered everything
3188s when it comes to this. I mean, there's
3190s little things like you can close and
3191s open the sidebar. You can add new
3193s sessions. You've got the set settings
3195s here, and you can choose how you group
3197s it by whether you want it manual or last
3198s updated and that sort of thing. You can
3200s also delete a project here. So, we can
3202s click on delete workspace and just
3203s remove that if we want to. And we have
3205s the ungrouped chats over here too as you
3207s can see. And [snorts] sometimes what you
3209s see here is like it gives you a local
3211s address like this. You can either open
3213s that up manually and just copy and paste
3214s that into your browser like that. Or the
3216s other option you have is you can just
3217s say you know open this up or turn this
3219s into a web page so I can view view it
3220s and preview it and that sort of thing
3222s which is quite nice. So these are the
3224s finishing touches and then you can see
3225s over here it's asking us for approval.
3227s So, we're going to click on allow. And
3229s that's beginning to edit the settings
3231s inside the DSH folder, which is where
3234s the visual UI of how it's being written
3237s is is right there. Right? So, the DSH,
3240s if you remember earlier, is the folder
3242s where the UI and everything that we
3244s build inside creator mode is is
3246s published. And you can see here, for
3247s example, it's editing the public/style
3250s CSS. And I think it's quite interesting
3252s as well because it sort of teaches you
3254s what it's doing and how it's coding, but
3256s you don't need to be technical to use
3258s it. So you can you can just keep it
3260s simple and give it simple like plain
3261s text prompts, but also you kind of learn
3264s like what it's doing and how it's coding
3265s out the project and where it's putting
3267s everything and why it's doing that and
3268s that sort of thing. And also if you
3270s click on these, you can see it's
3272s thinking. So you can see it's reasoning
3274s as it goes through, right? So it's like
3276s here's the CSS. Now let me consider this
3278s actually let me think about this. So, if
3280s you want to go like really in depth in
3282s terms of what it's doing, you can view
3283s over there. And now we can see we've got
3285s the new version of this setup as you can
3287s see, right? So, you can switch between
3290s um all of these different settings and
3292s it's just improved it slightly. I
3294s actually don't like the way that it's
3296s redone that, but yeah, you get the
3297s point. Like, you can just iterate and
3298s improve and stuff like that. One other
3300s thing that you can do here is you can
3301s build out skills. So, if we go to the
3303s settings, we can actually teach skills
3306s to our AI agents like our harness. So
3309s what does that mean? So basically like
3311s skills are also plugins inside the setup
3315s that we've got here. Right? So when
3317s we're for example like running SEO
3320s workflow, we can say okay save this as a
3322s plugin. Do that as a plugin. Right? So
3324s for example here, let's say I've got a
3326s skill for SEO or let's say for example
3329s we have a UI skill like UI UX Pro Max.
3332s We can grab that. Then we can go over to
3335s the aentic OS and we can say can you
3337s install this as a plugin as a skill so
3339s that you can actually create really nice
3341s UI. So if you see for example we can
3344s actually teach it as a skill and we can
3345s improve that directly there as well
3347s also. Pretty cool. So you know I was
3349s talking before about how to add deepsee
3352s v4 flash for free with open code. So, if
3355s we click on this dropown now and we
3356s click on the model, you can see that
3358s we've got Deep Seat V4 Flash, Deep Seat
3360s V4 Pro, and now we have added Deep Seat
3362s V4 Flash free. How cool is that? So,
3365s we've added a free model inside a
3366s harness. And that's just ready to go
3368s right there. So, if we actually look at
3370s what it's done over here, we said, okay,
3373s can you add open code as a drop down
3374s with Deep Seek V4 Flash so that I can
3376s use the harness for free when I'm
3377s building it? Said, sure, I'll
3379s investigate that. Now, let me build it.
3381s Um, and then it says the key harness.
3383s here's the information and then it
3384s started looking at what open code is. So
3386s it was like reading and thinking and
3388s having a look through our files. They
3389s said I've got the new uh the full page
3391s here. Right. So your harness ships the
3394s open code provider but it's mounted
3396s dormant. So it only appears in the model
3399s drop down once the setting profile
3400s activates it. So open code zen offers
3403s deep seat v4 flash for free. And so it
3405s needs an open code setup and then it's
3408s rooted that inside the settings so that
3410s it's a model that we can add on the
3412s dropdown. Right? So it says it ships
3414s with the drop down here and then we've
3417s got that plugged into the system. Pretty
3419s cool actually amazing that you can just
3421s change the drop down like that. I mean
3423s you could do the same for example if you
3424s wanted to code for free. You could also
3426s go to open router. You could go to omni
3428s router. You can actually type in free on
3431s open router here and you'll see all
3432s these different models that you can then
3434s plug in directly into deepseeek harness
3437s to code for free as well. So that's
3439s basically it. That is the full course.
3440s We've talked about all the different
3441s levels. The top level is aic operating
3443s systems. The basic level is just using
3446s like the coding functionality to build
3447s out whatever you want. We've covered
3449s skills, plugins, cordis, why it's so
3451s different, what makes it useful versus
3453s everyone else, and also [snorts] how to
3454s to get the most out of it and even add
3457s your own free models and how to plug it
3458s into an energetic operating system where
3460s you've got everything inside one place.
3462s Now, if you want to get more training on
3463s this sort of stuff, if you want to
3464s connect with me personally, feel free to
3466s join the AI profit boardroom link in the
3468s comments description or just go to the
3469s apiprofitboom.com. You see, we've got a
3471s bunch of nice reviews and testimonials
3473s right there. Inside the community, you
3474s can post questions, get help and support
3476s in real time, and I personally answer
3477s those questions on a daily basis. In the
3479s classroom, you can actually get access
3481s to all of our best trainings. So, we add
3483s new courses. If you like these sort of
3484s courses, we've got a bunch of courses
3485s over here. We have all my playbooks I
3487s personally use inside my own business.
3489s And then also inside the new daily
3491s update section, you can grab the Agentic
3493s operating system that we've built and
3494s we've just updated it. Plus, you can get
3496s the zip file to install it. And then you
3497s can get access to all of our new
3499s trainings and useful materials. Right
3501s here inside the calendar you can jump a
3502s weekly coaching call. So there are four
3504s weekly coaching calls per week where you
3506s can ask questions and get help and
3507s support in real time. And in the map you
3509s can meet people in your local area who
3510s are building out with AI agents like
3512s you. So I hope to see you inside there.
3513s Link in the comments description or just
3515s go to the arpebomb.com. That is the full
3517s Deepseek harness course as promised and
3521s excited to see what you build. She's
3523s watching.