0:01 What I'm about to show you is how I have
0:04 an army of AI agents working for me in
0:06 Discord 24 hours a day, seven days a
0:08 week, all autonomously running my
0:10 business and making me money. They're
0:11 building me apps. They're writing me
0:13 scripts. They're researching
0:15 competitors. They're finding trending
0:17 content. They're researching me stocks
0:20 to invest in an AI all autonomously, all
0:21 without any oversight needed, just
0:24 producing value non-stop. In this video,
0:25 I'm going to show you how to do the
0:28 exact same thing. I have been using
0:30 OpenClaw for over a 100 hours the last
0:32 month and I promise you this is the most
0:34 comprehensive multi- aent workflow
0:36 you'll ever see. When you integrate
0:38 OpenClaw into Discord, you're basically
0:41 able to create a multi- aent operating
0:43 system that is constantly producing
0:45 value for you autonomously. It's going
0:47 to save you time and potentially even
0:49 make you money. By the end of the video,
0:50 you'll have your entire multi- aent
0:53 workflow set up for you. And I'll also
0:55 answer all of these questions for you.
0:57 How do you set up OpenClaw and Discord?
0:59 How do you create a six agent system
1:00 that are all doing different things in
1:02 parallel? Which model should you be
1:04 using for each? Should you be using
1:07 local models? Can you even use local
1:08 models with your current device? How do
1:10 you make the agents proactive? How do
1:11 you set up a dashboard so you can
1:13 monitor your agents? How much would this
1:15 all cost? How do you improve your
1:18 security for your OpenC? And what use
1:20 cases can you set up today to get
1:22 maximum impact? That will all be
1:24 answered by the end of this video. Now,
1:26 let's lock in and get into it. So, it
1:27 turns out Discord has been the most
1:29 powerful way to actually interface with
1:31 OpenClaw out of all of them. The reason
1:33 being is you can set up these really
1:35 advanced workflows where you have
1:37 multiple channels set up and agents in
1:39 each channel doing different things for
1:41 you. This is impossible in Telegram or
1:43 iMessage or WhatsApp. It is only
1:45 possible in Discord. And be quite honest
1:46 with you, before this, I absolutely
1:47 hated Discord. I thought it was the
1:49 worst app ever. But I'm going to be
1:51 honest, I think Discord was built for
1:54 OpenClaw. I think Discord was made with
1:56 multi- aent setups in mind because this
1:58 is really amazing what I'm about to show
1:59 you. What I'm going to go through first
2:01 is my actual Discord setup, a bunch of
2:03 the workflows going through this, and
2:05 then I'm going to show you how to set it
2:06 up all yourself. I'm going to talk about
2:08 what's necessary, how to set it up, what
2:10 the cost would be, which models you
2:11 should use, all of that. So, let's first
2:13 quickly go through my workflows. There's
2:15 going to be a lot of different chapters
2:17 down below, so feel free to skip around
2:18 at any point and get to the parts you're
2:20 interested in, but let me go through my
2:22 workflow first. First, the structure my
2:23 Discord. There's a few different
2:25 sections here that are really important.
2:27 First is the normal text channels. These
2:29 are kind of where the automated flows
2:31 happen. When my agents do research, they
2:34 will drop off information in one of
2:36 these channels and then that will kick
2:38 off a workflow where another agent from
2:40 another channel grabs something. So, for
2:42 instance, check this out. In my alerts
2:45 channel, I have an agent that every two
2:47 hours goes and finds me tweets that are
2:49 starting to trend. These are tweets
2:51 specifically in my niche. So, vibe
2:55 coding, openclaw, AI. Anytime a tweet is
2:57 made that is starting to trend and pick
2:59 up, it brings me a list of those tweets
3:01 that are starting to trend. This kicks
3:03 off a really deep and complex workflow,
3:05 which I'll get into, but this is a
3:06 really important piece of information,
3:08 these trending tweets. The number one
3:11 way to go viral on X, is to talk about
3:13 things that are trending, is to talk
3:15 about the things that people care about.
3:16 And the only way you're going to know
3:17 what the people care about and what's
3:19 trending is if you get alerted when
3:21 things start to get on fire. The issue
3:23 is is that the main feed on X can get
3:24 crowded and busy and have a lot of
3:27 political slop. So, I like having this
3:29 alert system where the agent brings me
3:31 trending tweets. And again, I'll go over
3:34 how to set this up step by step by step
3:35 so you can do the exact same thing. But
3:37 just to show you how it all works, after
3:38 I get the alerts, I have another agent
3:41 come in and research the stories behind
3:42 the tweet. So, it goes, "Okay,
3:44 interesting. someone's talking about a
3:46 brand new anthropic model or a brand new
3:48 open AI model. It researches the story,
3:51 tells me what's going on, and gives me
3:53 different angles I can take on it. So,
3:55 what the interesting parts of each story
3:57 might be. Once Henry drops in these
4:00 research stories, I have another AI
4:03 agent, Quill, come in, take each one of
4:06 those stories, and then write me scripts
4:07 for each one of those stories. So,
4:09 YouTube scripts, potentially be tweets,
4:12 anything like that. It writes entire
4:14 scripts in my own voice based on my
4:16 other YouTube videos about those
4:19 stories. Then from there I can go in,
4:21 read the scripts and actually react. I
4:23 can say yes, this looks good and leave a
4:25 check mark or no, it doesn't look good
4:27 and leave an X based on my feedback. It
4:29 starts to train the model on what I like
4:30 and what I don't like. So the scripts I
4:32 get moving forward are more relevant to
4:34 me. Then from there, the moment I
4:36 approve a script, it actually spins up
4:39 another AI agent that gives me thumbnail
4:41 concepts for that video. So, I chose a
4:43 script I like and then another AI agent
4:45 spins up and gives me the exact
4:47 thumbnail concepts I could use. Now,
4:49 they're all in text. What I typically do
4:51 from this point is feed it into Gemini,
4:53 feed it into Nano Banana Pro, and get
4:55 the thumbnail made. Eventually, I'm
4:57 going to have a system set up, which
4:58 will be pretty easy, honestly, to hook
5:01 this into the Nano Banana API. So, it'll
5:03 just create the thumbnails for me and
5:04 drop them off here. But my other AI
5:06 agent pixel goes in and gets me these
5:09 thumbnails based on which scripts I
5:11 approve of. And this reduced my entire
5:15 content creation time from like 3 hours
5:17 of research down to like 5 minutes. It's
5:20 given me unbelievable leverage and freed
5:22 up a ton of time where I can go and do
5:24 other things I enjoy like tinkering and
5:25 building. But there's a whole lot of
5:26 other things going on in this world.
5:28 This isn't just for content creators.
5:29 There's a whole lot of other things
5:31 going on here as well. For instance, I
5:32 have a stock research channel. Every
5:35 morning at 7:00 a.m., I have an AI agent
5:38 spun up that goes and researches stocks
5:40 I'm interested in. So, I gave it a
5:42 description of what type of stocks I'm
5:43 looking for. I'm looking for companies
5:45 that are going to be involved with the
5:47 AI buildout over the next 10 years that
5:49 are creating hardware that's going to be
5:51 a bottleneck for the AI industry. So,
5:53 things like memory and GPUs and chips
5:55 and things like that. And every morning
5:58 I get a research report at 7:30 a.m.
6:00 that shows me the companies that are a
6:02 part of these bottlenecks and part of
6:04 the buildout and all the news on those
6:05 companies. This is like my stock
6:07 researcher agent and it's completely
6:10 automated. I used to spend hours a week
6:13 doing this research completely freed up
6:15 100% automated. Now I'm all caught up on
6:17 the news and what companies I should be
6:18 investing in. And this is totally
6:20 customizable. Maybe you don't care about
6:22 the AI buildout. That's totally fine.
6:24 And you can say, "Hey, research me
6:26 dividend stocks or research me financial
6:28 stocks, whatever you want. Research meme
6:30 stocks. It doesn't matter. You can
6:31 customize this any way you want." So,
6:34 you get automated stock research every
6:36 single morning. I have a competitor
6:38 research channel. So, what this does is
6:41 every morning at 7 a.m. as well, it goes
6:44 in, it finds me the top trending YouTube
6:46 videos on OpenClaw. I want to know the
6:48 moment any OpenClaw video on YouTube
6:50 goes viral. And this finds me every
6:53 video posted over the last 5 days that
6:55 are getting high amount of views per
6:57 hour. So we have 82 videos about
6:58 OpenClaw that have been posted the last
7:01 5 days on YouTube. And it ranks it in
7:05 order of views per hour. So I can go in
7:07 and see what kind of content is starting
7:10 to trend about Open Claw and Claude Code
7:12 and Vibe Coding and I can stay on top of
7:14 those trends and make sure I create
7:16 videos that are relevant to my viewers.
7:18 I even have an automated channel called
7:21 Daily Digest where it tells me every
7:23 single thing my agents have been doing
7:24 that day. So, what they've been focused
7:26 on, what they've created, what they're
7:28 waiting for me on, like what I have to
7:29 approve when it comes to scripts or
7:31 things to build out. It gives me an
7:33 entire rundown of everything they
7:36 completed that day. So, I know what my
7:38 agents are doing. And there's more to
7:40 it. I have direct lines to all my
7:42 agents. So, all my agents and sub
7:44 agents, you can see here, direct lines
7:46 to them. So I can open them up at any
7:48 time and talk to them directly. Charlie
7:50 is my coding agent. So this is my agent
7:53 just coding 24 hours a day. It is
7:55 constantly updating me on everything it
7:56 built. So it's in a Ralph loop,
7:58 constantly building things, creating
8:00 different features and functionality for
8:02 my SAS creator buddy, for different side
8:04 projects, for experimental projects, for
8:06 games, just so I can have fun, things
8:08 like that. It is constantly pinging me
8:10 in Discord with everything it's building
8:11 out so I can stay on top of it to make
8:13 sure it's productive. And then, and this
8:15 part's really important, I have a
8:17 section for all the major projects I'm
8:19 working on at the moment. So, what
8:21 you're probably doing at the moment is
8:22 you just have like one chat on Telegram
8:25 or WhatsApp where you just talk about
8:27 everything with your agent in that one
8:29 chat in that one channel. The issue with
8:30 that is is everything you're talking
8:32 about gets disorganized, right? Maybe it
8:34 creates documents, maybe it creates
8:36 artifacts, but now all those things are
8:38 just lost in that single stream chat.
8:40 Maybe you discuss an idea and then five
8:41 minutes later you come up with a new
8:43 idea and you jump over there and start
8:44 talking about that other idea and all of
8:45 a sudden everything you talked about
8:48 before gets lost. Not anymore. Now,
8:50 because I have channels for each project
8:53 and focus I have in my business, I'm
8:55 able to quickly go there and have a
8:57 complete history of everything I talked
8:59 about for that specific project. And
9:01 what it even does is it pins the
9:03 documents and research reports it
9:06 creates at the top of those channels. So
9:07 if you ever do something like ask it to
9:10 build an architecture diagram or come up
9:12 with a plan for you or something like
9:15 that now it's able to pin and organize
9:17 all those documents for you in one
9:19 place. Discord is such an amazing
9:22 operating system for your entire life
9:25 and it only becomes more powerful when
9:27 you mix in a multi- aent openclaw
9:29 workflow inside of it. So now I'm more
9:31 organized. I'm able to stay on top of my
9:33 agents that are actually building things
9:35 for me. And on top of it, I have all
9:36 these automated workflows that are
9:38 constantly producing value for me and
9:40 saving me tons of time. So, now that I
9:43 gave you a glimpse of what is possible,
9:45 let's talk about how to set this up. I'm
9:46 going to show you how to set a bunch of
9:48 the workflows I have that I just showed
9:50 you, it should give you a glimpse into
9:52 how you can set up any relevant
9:54 workflows that you have in your life.
9:55 Then, I'm going to answer a lot of the
9:57 other questions we talked about earlier,
9:58 like how much would this cost, what
9:59 model should you be using, things like
10:01 that. and then I'll go through some
10:03 exercises with you so you can find super
10:05 powerful workflows that are personal to
10:07 you that will help solve your specific
10:09 problems. So, let's go into the setup
10:10 itself. First of all, make sure you have
10:13 OpenClaw installed. I have entire videos
10:14 walking through getting OpenC installed.
10:16 So, if you don't have it installed,
10:17 check out a link down below for the
10:19 setup video. Once you do that, obviously
10:21 you need Discord installed, too. Inside
10:24 Discord, create your own private server.
10:26 So, mine is called Alex Finn Global
10:28 Enterprises. So, go in, create a new
10:30 server. Over on the lefth hand side, if
10:31 you scroll to the bottom, you just click
10:33 the plus sign and it creates a new
10:35 server for you. Give it a name. It's
10:36 just for you. It's not for anyone else.
10:38 Very critical. You do not let anyone
10:40 else come into your server. It is just
10:42 for you. Once you do that, you should
10:44 have a blank canvas inside your Discord.
10:46 No channels or anything. Now, we're
10:48 going to start setting this up with our
10:50 OpenClaw. The beautiful part about this
10:52 is your OpenClaw is able to handle a
10:54 vast majority of this setup itself. So,
10:56 all you need to do is say, "I want to
10:58 set you up in a Discord server so I can
11:00 communicate with you there. Please walk
11:02 me through getting you in as a bot in
11:04 Discord so I can communicate with you."
11:06 You hit enter on that and it's going to
11:07 start the process of getting it in
11:10 Discord. It's going to first add Discord
11:12 to its own config. Then from there, it's
11:14 going to walk through adding it as a bot
11:16 inside of Discord. This is the site it's
11:18 going to take you to, which is the
11:20 developer application site for Discord.
11:22 It's going to have you first create a
11:25 new application. This application is
11:26 going to be your bot. So, you're
11:28 basically going to be building a bot for
11:29 all your different AI agents. From
11:31 there, it's going to have you configure
11:33 different permissions. So, this will
11:34 walk you through it in your chat in
11:36 Telegram or WhatsApp or whatever you're
11:38 using and tell you exactly what
11:40 permissions to turn on. This should not
11:42 be complicated at all as long as you
11:43 follow precisely what your bot is saying
11:45 in Telegram step by step. So, it's going
11:47 to tell you which permissions to turn
11:49 on. It's going to tell you on this site
11:51 how to generate the token and then when
11:53 you generate the token on here, you give
11:55 it back to your bot and it's going to
11:57 initiate the bot for you inside of
11:59 Discord. Anytime you get confused or
12:01 anytime you don't know what's going on
12:04 or anytime it crashes, literally just go
12:06 to your bot and say, "Hey, this didn't
12:08 work for me or I'm not sure what to do
12:09 here. Where do I click?" It will tell
12:11 you exactly what to do. Never get
12:14 frustrated. Never get upset. anytime you
12:16 don't know what's happening. Just go to
12:17 your bot and say, "I don't know what's
12:19 happening. Help me out." And it will
12:20 tell you what to do. From there, your
12:22 bot will be connected and you will be
12:24 able to talk to it in Discord. So,
12:26 you'll probably have a channel directly
12:27 for your bot that you can talk to. Now,
12:29 you're in good shape. Now, you can start
12:31 setting up your different structure here
12:33 and your different workflows. The first
12:35 thing you probably want to set up that's
12:37 been the most helpful for me is the
12:38 different project channels. So, these
12:40 are a channel for every different
12:42 project you have. These are a few of the
12:44 things I'm focused on at the moment,
12:45 different things I'm building out. And
12:47 these are going to be projects where you
12:50 have like specific context for it, where
12:51 maybe you build a bunch of documents for
12:54 it or it's projects you come back to
12:56 quite often. You want channels
12:58 specifically for those projects. So,
12:59 here are some options for if you can't
13:01 think of any right now. Maybe you have
13:03 one for your career. Maybe you have one
13:05 for your personal life. You can have one
13:07 specifically for your Discord operating
13:09 system. If you're vibe coding and you're
13:12 building a ton of apps, make a channel
13:13 for each one of the apps you're building
13:15 out so you can come back and talk about
13:17 those things. What's incredible is in
13:19 each one of those channels that your bot
13:22 is in, it has full control over your
13:23 computer just like it normally would if
13:25 you're talking to it through WhatsApp or
13:26 Telegram. So you can say from these
13:28 channels, if you have like a vibe coding
13:29 channel, build this out from here, do
13:31 this for me. And it'll go and write code
13:33 on your computer just like it normally
13:35 would in Telegram. How do you set these
13:37 channels up? Really, really easily. You
13:39 come in, you say, "Hey, open claw or
13:40 whatever that your name is for your
13:42 claw. I want channels for each one of my
13:44 projects we are working on. Please build
13:46 out a channel in Discord for every major
13:48 project we are working on and make sure
13:50 I can communicate with you from each."
13:51 So, you can do it this way where you
13:53 kind of just lean on your OpenClaw to
13:55 figure out what the major projects
13:57 you're working on are. This is probably
13:58 the easiest way to do it. If you want to
14:00 be specific, you can do it that way as
14:02 well. But, I actually recommend this.
14:03 I'm going to put this down below for
14:05 you. that will set up your different
14:07 project channels for you, which is
14:09 really important. One of the core
14:10 components to everything I'm doing in
14:12 this Discord. It just allows you to be a
14:14 lot more organized than your single
14:16 stream chat that you have in Telegram or
14:17 WhatsApp. By the way, if you've learned
14:19 anything so far, make sure to leave a
14:21 like down below, subscribe, and turn on
14:23 notifications cuz all I do is create
14:25 amazing videos about AI and OpenClaw.
14:27 And make sure to check out the Vibe
14:29 Coding Academy. I do a live boot camp
14:31 every single week inside the Vibe Coding
14:33 Academy. You can come in, ask me
14:35 questions live, chat with me live. We
14:37 also have a ton of video series in there
14:38 as well. I go super in-depth on all
14:40 this. So, link for the Vibe Coding
14:42 Academy down below. Highly recommend it.
14:44 But next, let's talk about setting up
14:46 our advanced workflows like the ones I
14:47 showed you earlier around content and
14:49 research and all that. These can range
14:51 from simple to more advanced. I'll start
14:52 out with a couple of simple ones and get
14:54 your feet wet. Then, we'll get advanced
14:56 like that multi-channel content pipeline
14:58 I showed you earlier. I'm going to start
15:00 with the stock research automation. Even
15:02 if you're not interested in stocks, you
15:04 can use this to research stories or just
15:05 stay up to date on the news. So, this is
15:07 pretty important for everybody. What
15:09 you're going to want to do is go to your
15:12 direct agent line that was probably
15:13 originally set up. And then you're going
15:14 to want to describe what you want. So,
15:16 here's another prompt you can steal.
15:18 I'll put this down below as well. Please
15:20 build a new channel for me for stock
15:22 research. Every morning at 7 a.m.,
15:23 please build me a research report that
15:25 goes over important stocks involved with
15:27 the AI buildout. These should be stocks
15:29 that stand to benefit by AI being built
15:31 out more. They should also have
15:33 competitive advantages and strong modes.
15:35 These could be chips, energy, or any
15:37 other part of the supply chain. When you
15:39 hit send on that, your agent will go and
15:41 actually create these channels out. If
15:43 it doesn't have proper permissions for
15:45 this, it will ask you to go in and
15:46 change its permissioning so it can
15:48 create channels for you. And this will
15:49 do a few things. This will one create
15:52 the channel, but this will also two set
15:55 up a cron job for you. And basically
15:57 what this is is a scheduled task for you
15:59 every single morning. So I just added on
16:00 the end here, do this for 7 a.m. every
16:02 morning. And what's going to happen is
16:04 it will schedule at 7 a.m. every single
16:07 morning in perpetuity a sub agent that
16:10 it will spin up that will go and then
16:12 research online these exact
16:14 specifications. We put you know AI
16:16 companies that have modes and
16:17 competitive advantages. If you're not
16:19 interested in AI, that's fine. You can
16:20 change this to anything you want. If you
16:22 want to make this different crypto
16:23 coins, you can do that if you want. and
16:25 it'll research crypto coins for you.
16:26 Whatever you're interested in, it can do
16:28 that. If you're more just interested in
16:30 stories and news and you don't care
16:32 about stocks as much, you can say,
16:34 "Research all the latest news about
16:36 sports or whatever you're into," and it
16:38 will put it in here for you. And now you
16:40 have an automated research channel set
16:42 up. This is probably the simplest use
16:44 case you can set up. Let's get a little
16:45 bit more advanced, though. Let's talk
16:47 about doing content research or
16:49 competitive insights. So, I have this
16:50 competitor research channel that goes
16:52 and finds what my competitors are
16:54 posting about OpenCloth. And you can go
16:56 in and you can very simply say into your
16:58 chat in your main channel with your
17:00 agent. Please create a competitive
17:01 research channel for me that finds
17:04 YouTube videos from my competitors about
17:06 Vibe Coding and OpenClaw. Rank them by
17:07 trending. I want to find the newest
17:09 trending videos. Send it to me every
17:11 morning at 8:00 a.m. And this doesn't
17:13 even necessarily need to be competitive
17:14 research, right? This could just be if
17:16 you want to see what the latest Open
17:18 Claw videos are, you put that in there.
17:20 Or if you're more interested in sports,
17:21 find me the latest sports videos and put
17:23 in there. The latest basketball
17:24 highlights, whatever you need. This
17:26 isn't just for the specific use case I'm
17:28 showing. You can customize this any way
17:29 you want. And now this gets a little bit
17:31 more advanced because now what's going
17:33 to happen is it's probably going to ask
17:35 you for a YouTube API key. This is
17:37 completely free. Anyone can get this
17:39 quickly and put it in here. And what
17:42 that'll do is allow OpenClaw to actually
17:44 ping YouTube directly to get this video
17:47 information. It will walk you through
17:48 exactly how to do that. All you have to
17:50 do is go to a Google link, say generate
17:52 key, and copy and paste that in and
17:54 you'll be good to go. Your OpenClaw can
17:56 walk you through that much better than I
17:57 can because it'll know exactly what
17:59 you're doing on your screen. So again,
18:01 if you get stuck or confused at any
18:02 time, just say, "Hey, OpenClaw, I don't
18:03 know what's happening. Please help me
18:05 out." Don't be afraid to ask for help
18:07 from your OpenClaw. Now, let's get more
18:09 advanced. Let's talk about this multi-
18:11 channelannel workflow I got going on
18:13 where tweets are coming in from one
18:14 channel, then another channel is writing
18:15 scripts for me, and then another channel
18:17 is doing research. Let's build this out
18:18 step by step. So, you're going to go
18:21 back into your direct agent line with
18:23 your main agent here. Again, this prompt
18:25 will be down below. I want to set up a
18:27 multi- aent workflow in the Discord. I
18:29 want one channel where every morning an
18:31 agent gets spun up that researches
18:33 expost for trending content on OpenClaw.
18:35 Again, can be whatever you want. A half
18:37 hour after that, I want another sub
18:39 agent spun up that takes those tweets
18:41 and researches the stories behind them
18:43 and puts them in a research channel.
18:45 Then a half hour after that, have
18:46 another agent take those research
18:49 stories and create scripts for each that
18:51 I can approve that sends an indicator if
18:53 I like the script or not. This will have
18:55 your agent go and set up this
18:57 multi-channel system. It will set up
18:59 multiple channels here and then it will
19:02 also set up the cron jobs. What it's
19:04 also going to need from you is probably
19:07 a couple API keys. X is really locking
19:09 down their platform from bots. They do
19:11 not want bots crawling their site. And
19:13 so what you're going to need for this is
19:15 the X API key. If you're going to be
19:17 doing this based on exposts, you can get
19:20 an API key if you just search for X API.
19:22 It's pay as you go and it's pretty cheap
19:24 for every single tweet you pull. I'm not
19:25 spending that much money on it and I
19:26 think it's well worth it for the
19:28 research you can get out of it with
19:30 OpenCL. You plug that in and it will set
19:31 up its system where it will
19:33 automatically pull posts for you every
19:35 single morning. And your OpenClow will
19:36 also go and set up in those other
19:38 channels and schedule those other sub
19:40 agents to do each piece of this workflow
19:42 for you. Now, every morning you're going
19:44 to be woken up to pings of scripts being
19:46 written for you and content being
19:47 created for you. And even if you're not
19:49 a content creator, even if you're not
19:50 creating YouTube videos or posts or
19:52 whatever, you can still do different
19:54 things with this. One workflow I'm
19:55 setting up right now where you can steal
19:57 this from me and get a step ahead if
19:59 you'd like is actually taking these
20:01 research stories. So the same thing
20:02 happens. It takes the posts, then it
20:05 researches the stories, but instead of
20:07 writing YouTube scripts, it'll actually
20:10 build app prototypes for you. So imagine
20:12 this, your AI agents finding trending
20:14 stories and then finding challenges in
20:15 those stories and then just
20:18 automatically builds out prototypes for
20:20 those stories for you. Maybe there's a
20:22 trending tweet about Discord workflows.
20:24 Now, your agent will go and build a
20:26 prototype app out that maybe sets this
20:28 up automatically for people that you can
20:30 then turn around and go sell it to them.
20:32 Right now, there's new business lines
20:34 being set up for you. So, that's kind of
20:36 a even more advanced workflow I'm
20:37 currently setting up that you can steal
20:40 right now, which is doing research
20:42 online, finding trending stories, and
20:44 now you have a new channel where your AI
20:46 agents actually going and writing code
20:48 and building prototypes based on those
20:50 stories. That's my challenge to you to
20:51 set that up as well. From here, you can
20:53 do many things, right? You have a couple
20:56 automated workflows set up. Now, we can
20:58 do things like have direct agent lines.
21:00 I have a bunch of agents here. Some of
21:03 these are sub agents of Henry. Some of
21:05 these are actual full open claws. I
21:06 actually run three open claws because I
21:09 have two Mac Studios in a Mac Mini, but
21:10 you don't need to do that. But now you
21:12 can talk directly to your sub agent. So,
21:14 so if one of these sub agents like your
21:16 research sub agent like Echo for me, I
21:19 can go in and ask about specific stories
21:21 it wrote. Or maybe I have Quill who's
21:22 writing the scripts for me, I can go in
21:24 and give feedback about specific
21:27 scripts. You want to also set up direct
21:29 agent lines to each one of your agents
21:31 and sub agents as well. Again, as you
21:33 can guess, very easy. Go to your main
21:35 agent, go in and say, "I want to build
21:37 out direct agent line channels in this
21:39 Discord." And it will just set it up for
21:42 you. You will probably have to set up
21:44 new applications in Discord to set up
21:46 these individual bots. Again, it's gonna
21:47 be the same process you did before where
21:49 it walks you through the Discord site to
21:51 set up those bots. So, which model
21:53 should you be using for all this and how
21:54 much is this going to cost you? So,
21:56 there's two things you need to think
21:58 about. The muscles and the brains of
22:00 this operation. The brain's going to be
22:01 the orchestrator. This is the agent
22:03 you're talking to to build all this out.
22:05 And the muscles are going to be the sub
22:07 agents or the other agents that do the
22:09 dirty work. The best brain to use for
22:11 all this, the best orchestrator for
22:13 everything that's going on here is
22:15 Anthropic. It just is. Even though
22:17 they're doing a a few anti-consumer
22:19 things lately, cutting people off,
22:21 banning people, whatever it is, they
22:23 have the best model for openclaw brain.
22:25 It just is what it is. So, if you are
22:27 still on anthropic oath, I would use
22:29 that for the brain. I'm hearing from
22:31 some people, they're not able to set up
22:33 the OOTH anymore. It just simply does
22:35 not let them. If that is the case for
22:37 you, I'd highly recommend going to
22:40 ChatGpt. The Chat GBT models are still
22:42 crazy strong, crazy smart, not as
22:45 personable and warm as the anthropic
22:46 ones, but that's fine. They're still
22:48 very, very smart. And on top of that,
22:51 OpenAI is being very proconsumer and
22:53 encouraging you to use their OOTH with
22:55 Open Claw. So, you can very easily just
22:58 sign up for a $20 a month plan and use
23:00 Chad GBT. And thrive is going to cost
23:02 you $200 a month if you want full usage
23:03 out of it. So, it's the most expensive,
23:05 but it's the best. If you can't use it,
23:07 Chad GBT, you could probably get away
23:09 with the $20 plan, and if you need to
23:11 upgrade from there, if you really want
23:13 to save money, some of these cheaper
23:15 Chinese models are very strong as well
23:18 for a very good price. Kim K 2.5 is
23:20 probably the smartest Chinese model, but
23:21 you can also save a ton of money by
23:24 using Miniax 2.5. Those are probably the
23:26 two best Chinese models, and those are
23:28 going to be your best brains. Then
23:29 you're going to want to think about the
23:31 muscles. So the models actually doing
23:34 the work. From there, I would highly
23:36 recommend going cheaper. So going with
23:37 the cheaper Chinese models for that
23:38 because they're going to be doing a lot
23:40 of the dirty work. And as long as
23:42 there's a smart model overlooking it,
23:44 you don't need to get genius level for
23:46 the dirty work cuz the boss and the
23:47 manager is going to make sure it does
23:49 good work. The best option though and
23:51 what I think is the future of all of
23:54 this is local models. So, I have
23:57 Anthropic as my brain orchestrating
23:59 everything through the API, but then I
24:01 have local models doing all the dirty
24:03 work. And for those unaware, local
24:05 models are just AI models that don't run
24:07 in the cloud. We don't use an API to
24:09 talk to them. They run locally on device
24:11 on your own computer. I have three Mac
24:14 Studios in a Mac Mini just running local
24:16 models all day. This allows me to do a
24:19 lot more work basically for free just
24:20 for the cost of the energy. And that's
24:22 the advantage local models are going to
24:24 be able to give you is that they can
24:27 just run 247 365. No rate limits, no
24:29 cost. They just keep going and going and
24:30 going so you can get a lot of work done.
24:33 It just requires the cost of the
24:35 compute. So the cost of the Mac Studio.
24:37 You don't need to do this though. I
24:39 don't recommend just rushing into doing
24:42 this. I recommend using cloud API,
24:43 getting used to it, figuring out your
24:45 workflows, and then when you're ready
24:47 and want to make the jump to the local
24:49 world, you can then buy a computer from
24:51 there. You can still run local models
24:52 even if you're on a Mac Mini. They're
24:54 not going to be as smart, efficient, or
24:56 fast, but it's still cool to jump into
24:58 and have a local model running. A lot of
25:00 times, even if you're on the Mac Mini 16
25:02 GB, which is the cheapest one, you can
25:05 download Gemma, which is like a really,
25:08 really small cheap model that can manage
25:09 your memory for you. So, there's still
25:11 local models you can run even on Mac
25:13 minis. I'd ask your open claw about
25:14 it'll tell you exactly which Gemma model
25:16 to download and how to install it. So, I
25:18 showed you the models, which ones are
25:19 expensive, which ones are cheaper, and
25:21 what I think the end state would be. I
25:23 honestly think in the next 5 years,
25:25 we're all going to have compute on our
25:27 desk running models locally. So, I'm
25:29 preparing for that future now by buying
25:31 all these Mac Studios, but I think it'll
25:32 only get cheaper and easier over time.
25:33 And to be honest with you, you don't
25:34 need to rush into it if you don't want
25:36 to. Let's talk about devices you're
25:38 going to need for this. First of all,
25:40 you don't need to run out and buy a Mac
25:42 Mini or a Mac Studio. Just use any
25:45 device. Use anything except for a VPS.
25:47 VPS's will make this very difficult.
25:49 VPS's you're going to have to pay a lot
25:51 of money just to scale up to have the
25:53 memory to run multiple agents at one
25:56 time. I think VPS's are by far the worst
25:58 option for running OpenClaw in any
26:00 context or any capacity. Do not use a
26:02 VPS. Just have any device. I don't care
26:05 if it's an old crappy laptop you have.
26:06 Just take it out and run it. It's going
26:08 to give you a much better experience of
26:10 running OpenClaw. If you want to start
26:12 scaling up, the Mac Mini, I think, is
26:15 the best value in computing, bar none.
26:18 For only $600, you get an incredible
26:19 computer with one of the most powerful
26:22 chips in the world, the M4 or the M5
26:24 coming soon. You get tons of memory and
26:25 storage. You're in a really good spot.
26:27 If you have the cheapest Mac Mini there
26:29 is, you can do this entire workflow I
26:31 showed you. And if you want to get
26:32 really advanced, the Mac Studio, which
26:35 is what I have, allows you to run the
26:37 local models. So you can combine the
26:38 cloud models and the local models to do
26:40 a whole bunch of really amazing
26:42 workflows at once. These are my
26:44 recommendations. Basically anything
26:47 except for a VPS is what I recommend. So
26:49 some security measures, basically the
26:52 crux of all security advice I can give
26:54 when it comes to this workflow is not
26:56 letting anyone get involved with this
26:58 workflow. What does that mean? Do not
27:00 let anyone into your Discord server.
27:02 Don't let friends, family, anything into
27:04 this Discord server. Your agent has
27:06 complete access to your entire digital
27:08 life. If someone gets access to your
27:10 Discord server, they basically have
27:12 control over your entire digital life.
27:13 They can do anything at all. Do not let
27:15 anyone else into your Discord server.
27:17 Don't put these bots in other Discord
27:20 servers. Don't give other people access.
27:22 Keep this completely private to
27:25 yourself. And do not make workflows
27:27 where these agents have any sort of
27:28 right access to your digital
27:30 communication channels. You don't want
27:32 it spinning up emails on the fly
27:34 automatically and autonomously. You
27:35 don't want to have it sending text
27:37 messages. There's a lot of risks with
27:39 that. You just don't want to get that
27:42 involved. So, keep it personal. Keep it
27:43 private. Don't let anyone else get
27:45 involved. Have it doing work locally on
27:46 your computer. Don't have it
27:48 autonomously sending out emails and
27:49 messages just yet. These are going to be
27:51 the biggest piece of security advice I
27:53 can give. And also, always stay up
27:56 todate on your OpenClaw versions. They
27:59 are constantly adding in new security
28:01 features. So stay up to date on your
28:04 OpenClaw. Next, let's talk about finding
28:06 your own use cases. I showed you my use
28:08 cases. I showed you how to set it up. I
28:10 showed you everything you need to do for
28:12 my use cases. But this is the most
28:14 important part. This is about finding
28:17 custom use cases just for you. I make a
28:18 lot of tutorial videos. I show you
28:20 everything I do. I show you how to set
28:22 it all up. But the thing is, if you just
28:24 copy everything I do, you're not going
28:26 to get the most out of OpenClaw.
28:29 OpenClaw is the most incredible personal
28:31 operating system in the entire world.
28:33 And if all you do is just copy me,
28:34 you're not going to get the most out of
28:36 it. You want use cases that are custom
28:38 for you. But how do you find use cases
28:39 that are custom for you? Well, if you've
28:41 seen a lot of my videos before, this
28:43 will not come as surprise to you, but
28:45 it's still the most important thing in
28:47 the world, and that is reverse
28:49 prompting. Steal this reverse prompt.
28:51 based on everything you know about me,
28:52 my goals, my ambitions, and workflows
28:54 we've done in the past. What are some
28:57 advanced multi- aent automations we can
28:59 create in Discord? You hit enter on
29:01 that. That's a reverse prompt. That's
29:03 you asking your AI what you can do.
29:05 That's you asking your AI what is
29:06 possible. That's going to come up with
29:08 some incredible workflows for you to
29:10 implement in Discord. You choose a
29:12 couple of those workflows. You have it
29:14 implemented. It does it for you. You're
29:16 all set up and you're good to go. Steal
29:17 that reverse prompt. I put it down
29:19 below. It's the most important part of
29:20 the video. It's the most important thing
29:23 you can do because it will find custom
29:25 workflows for you. And that should get
29:27 the spark going in your head of other
29:29 things you can set up and do. Very
29:31 critical. Do that immediately. As you do
29:33 this and as you go, your bot's going to
29:34 come up with different ways to set up
29:37 these complex systems with multiple AI
29:39 agents using different models, things
29:41 like that. So for instance, even though
29:44 anthropic is my brain, it is spinning up
29:47 codecs and the chat GPT API to do some
29:48 of the advanced coding work. It's using
29:51 my local models for other. So make sure
29:53 when you do this, you ask, what are ways
29:56 we can use multiple agents to do this?
29:57 What are some ways you can use sub
29:59 agents and other agents to do a lot of
30:02 these workflows we set up? And it will
30:04 start spinning up these other agents
30:06 that will be doing the work for you. It
30:08 will create the channels, have you
30:09 create the bots, and you will now have
30:11 multiple agents doing a lot of this
30:14 work. It really is dependent though on
30:16 the workflows you're doing, right? So,
30:18 Quill is my script writer. You might not
30:20 need a script writing AI agent. It would
30:21 be silly for you to set up a script
30:23 writing AI agent if you're not writing
30:25 YouTube scripts, right? So, depending on
30:27 what workflows you come up with from
30:30 that last reverse prompt, ask which sub
30:32 agents would be helpful for your main
30:34 agent to spin up to do this extra work.
30:36 The next thing I want to talk about is
30:38 mission control and your custom
30:39 dashboard that you can make. So that
30:40 again, this is a little bit more
30:42 advanced. I'll also say this, I've been
30:44 teasing it for a while and people in the
30:47 comments keep demanding it. I will do a
30:50 full deep dive mission control video
30:52 next. That will be my next video unless
30:54 some breaking news happens. I have to
30:55 talk about the breaking news. But my
30:57 next openclaw deep dive video will be
31:00 mission control. You can also make an
31:02 agent dashboard inside your mission
31:04 control to monitor what all your agents
31:07 are doing in Discord. As you can see
31:09 here, I see all my agents I have working
31:10 at the moment. Most of them are idle.
31:12 Violet is working on doing some research
31:15 on my Mac Mini off a local model, but
31:17 you can see a live activity feed of
31:19 everything going on here. And this is a
31:21 really good way to track what your
31:23 agents are doing, what they're up to,
31:25 see the amount of events going on, see
31:27 who's the most active agent, understand
31:28 the amount of tasks completed, so you
31:30 understand their efficiency. This is all
31:32 in my mission control. All you need to
31:34 do to set this up is you go to your open
31:36 claw and you say, "Hey, I want a
31:38 dashboard that you can build that shows
31:41 me every agent working inside of mission
31:43 control. I want to see every agent in
31:45 our Discord. I want to see the tasks
31:46 they're doing, and I want an activity
31:49 feed. please build this for me. Your
31:51 agent will go and build this out. It
31:52 will launch it for you and walk you
31:54 through how to do it and you can
31:55 completely customize it. You want to
31:56 make it look different, go to your
31:58 agent, say, "Make it look different."
31:59 You want to add an activity feed, say,
32:01 "Hey, please build an activity feed." It
32:02 will vibe code everything for you. This
32:04 is not complicated. This is not complex.
32:06 If you want something, ask your agent
32:08 for it. Doesn't matter if you don't know
32:09 how to do it. Your agent will figure it
32:11 out. It's super intelligent. So,
32:14 building an agent dashboard inside your
32:16 mission control very, very cool as well.
32:18 If you want the full mission control
32:20 deep dive, make sure to subscribe and
32:22 turn on notifications. So the moment I
32:25 do it, it's available. I also did an
32:27 entire boot camp on mission control last
32:30 week, a 2hour mission control boot camp
32:32 inside the Vibe Coding Academy. You sign
32:34 up now, you get instant access to that.
32:36 But that was the my entire automated
32:38 Discord workflow. I showed you all the
32:40 channels. I showed you all the automated
32:42 workflows. I showed you all the
32:44 projects. Everything how it was set up.
32:46 It is the most powerful way to build a
32:49 multi- aent automated workflow in the
32:51 entire game. It is incredible. And my
32:53 entire workflow all goes through Discord
32:54 now, which I never thought I'd be saying
32:56 because before this, I literally hated
32:58 Discord. I hope this was helpful. Let me
33:00 know down below what you want my next
33:02 video about OpenClaw to be about. It's
33:04 going to be mission control, but I'll
33:05 read the comments and make the next one
33:06 about whatever the most requested
33:08 comment is. Hope this was helpful. I
33:10 absolutely love making these videos for
33:12 you guys. It is the most fun part of my
33:14 week. Thank you for all your support and
