Master 96% of Codex in under 35 minutes — Transcript
Full transcript
- 0:00Today I'm going over the 18 core Codeex
- 0:02concepts that you actually need to know
- 0:03in order to start using it right away
- 0:05and getting value from it. It doesn't
- 0:06matter if you're not technical at all or
- 0:07you've never used Codex before. By the
- 0:09end of this video, you'll understand
- 0:10exactly how this thing actually works so
- 0:11that you can start using it right away.
- 0:13So, let's not waste any time and just
- 0:14get straight into this one. All right,
- 0:15so I've split these 18 core concepts
- 0:16into four different parts and they get
- 0:18cooler as we continue to go on. So, part
- 0:20one, we're going to start off here is
- 0:22foundations. So, we're going to kick off
- 0:23with concept number one, which is
- 0:25projects. Now, a lot of you guys, I'm
- 0:27assuming, have used chatgbt or claude in
- 0:28the web before where you can talk to an
- 0:30LLM and it can, you know, give you an
- 0:32answer. But the problem you face is that
- 0:34every time you're talking to it or you
- 0:35start a new chat, you have to kind of
- 0:37familiarize it with you and your
- 0:38business and what you're working on that
- 0:40week or that quarter. So, with projects,
- 0:41you can basically keep everything
- 0:42organized into one project so that as
- 0:45you keep working inside of it, it only
- 0:46gets smarter about knowing who you are.
- 0:48So, right here on the lefth hand side,
- 0:49you can see that I've got a few projects
- 0:50going on. I've got Herk 2, I've got
- 0:52Hyperframes, I've got Trading Challenge,
- 0:54I've got AIS demo, and I typically am
- 0:56always just working inside of my Herk 2
- 0:57project. My Herk 2 project is basically
- 0:59what I call my AI operating system. And
- 1:01what a project is is actually very
- 1:02simple. It is just a collection of
- 1:04folders and files. So right here, this
- 1:06is inside of my file explorer on my
- 1:07computer and it's called Herk 2. And
- 1:09this is exactly the project that I have
- 1:11open in Codeex. When I say right here,
- 1:13I'm inside of Codex, right? This is the
- 1:15file path that I'm working in, which is
- 1:16if I open this up, my Herk 2 project. So
- 1:19what that means is this project has
- 1:21rules about me. It has, you know,
- 1:22credentials. It has all my projects. It
- 1:25has all my YouTube videos. This thing
- 1:26has so much knowledge about me. So now
- 1:28when I come into the chat and I say,
- 1:30"Hey, can you just give me a rundown of
- 1:32what we've been working on this past
- 1:34month?" This thing basically gets
- 1:35familiarized with my project and then it
- 1:37looks at everything that we've been
- 1:38working on together. It even looks
- 1:40through all of my previous chats that
- 1:41I've had with Codeex. So here you can
- 1:43see that it looked back from the past
- 1:44month and it said, "Here's what we've
- 1:46worked on. We've been improving your
- 1:47Herk 2 and your AISOS. We've been
- 1:50working on YouTube content and strategy,
- 1:51AI model testing and demos, apps and
- 1:53product experiments, websites, branding,
- 1:56business operations. So, this is no
- 1:58longer just an AI chatbot. This is now a
- 2:00co-founder. This is a personal
- 2:02assistant, which brings us on really
- 2:03nicely into concept number two, which is
- 2:05something called an agents.md.
- 2:08And don't let that MD intimidate you
- 2:10here. If I go over to the right hand
- 2:11side and I open up my files, you'll
- 2:13notice if I scroll down, I have
- 2:15something right here called an
- 2:15agents.mmd. And when I open this up,
- 2:18it's literally just a markdown file.
- 2:20That's what the MD means. And it's just
- 2:22rules. Like this is a very simple thing
- 2:23that you all can read and it's not
- 2:25technical at all. So this is basically
- 2:27like the rules for your project. So you
- 2:29remember inside of my Herk 2 project, I
- 2:31have this agents onMD. Well, if I was
- 2:32inside of my Hyperframes editor project,
- 2:34there would be a different agents.m MD.
- 2:36So this thing basically just sets the
- 2:37ground rules for the project that you're
- 2:39working in. So right here you can see
- 2:40that this one says, "Hey, you are Nate
- 2:41Herk's AI operating system. Your job is
- 2:43to help him spend less time on
- 2:44operations so he can focus on making
- 2:46YouTube videos." And then I give it
- 2:47things like core operating rules for how
- 2:49to work with me. I give it some stuff
- 2:50about how to be safe on my desktop. And
- 2:52then the real bulk of my agent MD is
- 2:54what I call routing. So this is a
- 2:56routing map. It basically means, okay,
- 2:58so yeah, Nate's project is huge. There
- 3:00are hundreds, probably thousands of
- 3:02files and folders inside of this
- 3:03project. So let me just explain to you
- 3:05how this actually works. If you need
- 3:07business stuff, you go to the wiki. If
- 3:09you need corporate structure, you go to
- 3:10the corporate structure section inside
- 3:11the wiki. If you need his voice, you go
- 3:13here. If you need course knowledge, you
- 3:14go here. If you want active projects,
- 3:15you go here. So everything that I'm
- 3:17doing, we're routing it back somewhere
- 3:19so that my agents.mmd file tells this
- 3:22project where everything is so that
- 3:24every time I open up a new chat, like I
- 3:25said, I don't have to reexplain things.
- 3:27And basically the way that this works,
- 3:28if we picture this being a chat thread,
- 3:30you know, you come over here and you
- 3:31say, "Hey, you say, "Hey, Codex, I need
- 3:33you to help me do this." And before
- 3:35Codex actually even like reads this
- 3:37message, what it does is it first reads
- 3:39the agents.mmd. So I'm just going to put
- 3:41a here and that stands for agents.mmd.
- 3:43But codeex will basically read this and
- 3:45then it will read your message and then
- 3:46it will respond with something like, you
- 3:48know, hey, how can I help you today? So
- 3:49the kind of stuff that you want to put
- 3:50inside your agents.mmd is the type of
- 3:52stuff that you want codeex to know about
- 3:54you every single time you're talking to
- 3:56it in that project. And now moving on to
- 3:58concept number three, we have something
- 4:00called the agent loop. So you'll hear
- 4:02about something like codecs or claude
- 4:04code or Hermes agent. These are all
- 4:07called agent harnesses. And these
- 4:09harnesses have their own agent loops.
- 4:11And real quick, if you're wondering
- 4:12about codecs versus like work, chatbt
- 4:15work, I always use codecs. The majority
- 4:17of my work is not building products or
- 4:19building software, but I only use codecs
- 4:21because it's just the most powerful. And
- 4:22once you get through this video, you'll
- 4:23realize how easy it is to use. But
- 4:25anyways, the reason why this is called a
- 4:26harness is because they have this thing
- 4:28called the agentic loop, which at the
- 4:30highest level basically just means when
- 4:31you ask the agent a question, it has a
- 4:34bunch of tools at its disposal and it
- 4:36basically thinks and it reasons and it
- 4:38uses tools and then it responds to you
- 4:40and then it thinks and then it reasons
- 4:41and it takes action. It uses tools,
- 4:43responds to you and that's basically the
- 4:44agent loop. And what's really awesome
- 4:46about that is you can see it in action.
- 4:48So right here you can see that this said
- 4:49worked for 25 seconds. And if I open
- 4:51this up, we can basically see what it
- 4:53did, right? So like it hid this response
- 4:55and it gave us our output, but it had a
- 4:57reasoning loop inside. It first said,
- 4:59"Okay, I'll check our recent tasks and
- 5:00project nodes, then I'll pull together
- 5:02the main themes from the past month."
- 5:03And when it did that, it ran all these
- 5:05commands. It ran this, which you don't
- 5:07have to know what this means, but
- 5:08basically what this did is it looked
- 5:10inside something right here called a
- 5:11memory.md. So very similar to the
- 5:13agents.md, but this is about memories
- 5:15rather than just like rules. It then
- 5:18looked at other chats. And so we can
- 5:19basically watch it do things. So, to
- 5:21show you guys a real example of this,
- 5:23hey, could you just check my most recent
- 5:24YouTube video that I uploaded and give
- 5:26me three comments that you thought were
- 5:28funny from that video.
- 5:31Now, as I shoot this off, what I want
- 5:32you guys to pay attention to is what
- 5:34we're watching. So, first it said
- 5:35thinking. Now, it says, "I'll find your
- 5:36latest upload, read through the
- 5:38comments, and pick three that made me
- 5:39laugh." And now we can see what it's
- 5:40doing. So, it's looking through the
- 5:42memory. It's looking at these things.
- 5:44It's running these commands. And these
- 5:45commands are basically right here. It's
- 5:47reading, you know, files to get
- 5:49established and to get oriented with
- 5:51what it needs to do. And so, you can
- 5:52actually learn a lot about how codeex
- 5:55works and how to best work with it by
- 5:57when you ask it a question, you just
- 5:58kind of watch what it does. So, you can
- 6:00see what it had to do was it had to run
- 6:01a command in order to actually go talk
- 6:04to YouTube, look at the data, and pull
- 6:06it back. And now we can see three
- 6:08comments. We have this one, never seen
- 6:10anyone so excited to lose money. Trading
- 6:12with AI is very easy. I just prompt
- 6:13Claude only take profitable trades. And
- 6:15if I want to lose money, I can do it
- 6:17myself faster and better. And what
- 6:18you'll notice is how quick that all
- 6:20happened. Now, I do have my Astra on
- 6:21fast mode, but I will talk about that
- 6:23later because that is a core concept
- 6:24that's coming up. All right, and moving
- 6:26on to the last concept of part one. It's
- 6:29a slashgoal. And this is honestly one of
- 6:32my favorite things about AI ever. It's
- 6:34the ability to do something called a
- 6:36slash goal. And right here, you can see
- 6:37that that says set a goal to keep
- 6:39pursuing. Now, AI agents are super goal
- 6:41oriented, which is super cool. So, let's
- 6:43say you have Mr. AI agent right here and
- 6:45you basically say, "Hey, here is your
- 6:47goal. You know, this is what you need to
- 6:49do." And you give it this objective
- 6:50goal. What happens is it will basically
- 6:52just keep working and keep working and
- 6:54keep working until it actually hits that
- 6:57goal. And it's cool because even if you
- 6:58kind of like give them a roadblock, they
- 7:00are going to be so goal-oriented that
- 7:03they're going to keep retrying different
- 7:04things and they're going to just
- 7:06basically keep going until they're able
- 7:08to succeed and hit that goal for you.
- 7:10Now, obviously, the more objective a
- 7:11goal, the easier it is to prove that.
- 7:13So, if you say, "Hey, can you, you know,
- 7:15work on this until you've pulled in 257
- 7:18sources and then write the report."
- 7:20That's an objective goal, right? 257
- 7:22sources and written report. But you can
- 7:24also set goals that are a little bit
- 7:25more emotional. Like your goal could
- 7:27maybe be more like, you know, until
- 7:29you're fully satisfied or until you've,
- 7:31you know, verified this over and over
- 7:33and you feel good about it. And in that
- 7:34case, you're still going to, you know,
- 7:37get there and it will still decide when
- 7:39it's done. But as you get more objective
- 7:41with your goal, the goal prompts are
- 7:43just more effective. But let me show you
- 7:45one goal that I actually did that you
- 7:46guys have already been witnessing in
- 7:48this specific example is I went to my
- 7:50hyperframes project. And hyperframes
- 7:51basically lets your coding agents build
- 7:53HTML and then render it as video. So
- 7:55essentially it creates video. But what I
- 7:57did here is I shot it off a/goal prompt.
- 7:59You can see this one says sent as goal.
- 8:01So basically what I did is I said I need
- 8:03you to help me create some motion
- 8:04graphics for a YouTube video. So all the
- 8:05motion graphics you've seen so far were
- 8:07from this exact prompt right here. I'm
- 8:09going over 18 different Codex concepts.
- 8:11I want, you know, an intro. I want
- 8:12transition cards. I want all this. So, I
- 8:14basically defined that I wanted one
- 8:15intro scene and that I wanted 18 cards.
- 8:18And then you can see that it said I
- 8:19finished and I visually checked, meaning
- 8:20it actually took screenshots to make
- 8:22sure everything looked good before it
- 8:23decided to be done. And now we have all
- 8:25these animations. But what's cool about
- 8:27the goal prompts is that models nowadays
- 8:29are getting so good, the AI models are
- 8:31getting so good that it's actually
- 8:33better to give them a goal and then step
- 8:35out of their way. If you really just try
- 8:36to like micromanage them, you're kind of
- 8:38just like muzzling their capabilities.
- 8:40So, if the situation isn't super super
- 8:42risky, give them a goal and you can get
- 8:43out of the way. Now, there's one quick
- 8:45thing that I felt the need to address,
- 8:46which is basically the question, "But
- 8:48Nate, what if we've been using cloud
- 8:49code for a long time?" Well, I'm really
- 8:51glad that you asked. It's actually
- 8:52really simple. Really, the main
- 8:54difference is the agents.mmd, thecodex,
- 8:56and the aagents. Because in cloud code,
- 8:59you basically have like your claw.md and
- 9:01you have yourclaw. So when I first
- 9:02started switching over from cla code to
- 9:04codeex. All I did was I said, "Hey
- 9:05codeex, take a look at this herku
- 9:07project, I've been building this up
- 9:09using claude code for the past couple
- 9:10months. I need you to help me get this
- 9:12codeex ready." And basically what that
- 9:14means is I made a copy of my claw.md and
- 9:16called it agents.mmd.d. So I have two of
- 9:18those now. One one agents.mmd one
- 9:19claude, but they're basically the same
- 9:21file. And then I made a copy of all my
- 9:22claude skills and put them in the agents
- 9:24folder. And I made a copy of all of my
- 9:26like claude settings files and I put
- 9:28them in my agents. And the cool thing is
- 9:30you don't have to do any of that
- 9:31manually. You'd literally just say,
- 9:32"Hey, Codeex, look at the documentation,
- 9:34analyze my project, and make this codeex
- 9:36accessible. Make it codec ready." So,
- 9:38anyways, if you were feeling a little
- 9:39bit of doubt about all that kind of
- 9:41stuff, that's the way I did it. And now
- 9:43you have a bunch of local files and
- 9:44folders that you can use on any agent
- 9:46harness whenever. It's very flexible.
- 9:48It's very tool agnostic, and that's what
- 9:49you want to be building at the end of
- 9:50the day. So, anyways, had to address
- 9:52that real quick. Let's get back to the
- 9:53video. All right. Awesome. So, let's
- 9:54move on to part two of these core
- 9:56concepts, which is environments. So the
- 9:59first piece of environments is number
- 10:00five, which is local versus cloud. So
- 10:03there's a couple things to talk about
- 10:04here. The first thing is if someone
- 10:06says, "Oh, is this running locally or
- 10:08are these local files?" All that means
- 10:10is does that exist only on the machine
- 10:12that you're currently using? Whether
- 10:13you're on a MacBook right now or you're
- 10:15on a desktop PC like I am right now.
- 10:16When you say local, it just means it's
- 10:18only accessible by you. So right here,
- 10:20if I open up my files, these are
- 10:22basically local files. Now, they're
- 10:23syncing with one drive when I want them
- 10:25to. So that would kind of push them into
- 10:26the cloud, but otherwise it's completely
- 10:28local. So like if you're working on a
- 10:29Word doc on your computer and you forgot
- 10:32to save it and then email it to yourself
- 10:34on so you could work on it on a
- 10:36different laptop, that was a problem
- 10:37because it was locally on that other
- 10:39machine. But something like Google Drive
- 10:40is always in the cloud obviously. But
- 10:42this also goes beyond just files. This
- 10:44goes to like things that are produced.
- 10:46So right here you can see that it
- 10:47actually created me this little site for
- 10:49all these motion graphics. So if I open
- 10:50up this site, you can see that I can
- 10:52view all of these videos. I can click
- 10:54into the different ones and it threw
- 10:55this together for me really nice and it
- 10:57served this to me on a local host. So
- 10:59right here you can see that this URL is
- 11:01127.0.0.1
- 11:03and the port of this local host is 8008.
- 11:06Now realistically that's not super
- 11:08important. You don't really need to
- 11:09understand what exactly this means. But
- 11:11this is just a local host. But this is a
- 11:14little confusing because it looks like
- 11:15this is a web address that if I, you
- 11:16know, emailed to you, you could open
- 11:18this up and have the site. But what
- 11:19would happen is this would show on your
- 11:21screen or it would show on your computer
- 11:23as nothing actually lives here. So like
- 11:25if I copied this and I pasted this into
- 11:26a new browser but I changed the number
- 11:28to 7. I don't think I have anything
- 11:31running on port 8087
- 11:34on my local machine even though you
- 11:35might. So nothing actually loads up. And
- 11:37it's quite funny. I remember seeing
- 11:38these these tweets where I think it was
- 11:40a joke but it was like uh someone was
- 11:42pretending they were a beginner and they
- 11:44were like wow cloud code and codeex it's
- 11:46so cool. Look what I built in one day.
- 11:47and then they, you know, attached a
- 11:49bunch of local host addresses and
- 11:51obviously no one else could open that up
- 11:53because they're not on that local
- 11:54device. So anyways, that is the
- 11:55difference between local and cloud. Now,
- 11:57the other piece here is that when you're
- 11:58spinning up these new chats, you can
- 12:00kind of have these be either local,
- 12:02meaning right here, you know, I'm
- 12:03working in my local files and folders.
- 12:05And that's great because codeex can
- 12:07actually like edit these files, right?
- 12:09It can delete things, it can make new
- 12:10ones, it can edit them, it can move them
- 12:12around, it can go to my downloads, it
- 12:14can go to my desktop, it can do anything
- 12:15on my computer really. or you could
- 12:17actually have these work in the cloud.
- 12:18And what you'll notice here is when you
- 12:19choose cloud, you no longer have the
- 12:21ability down here to actually choose the
- 12:23model you were using and the permissions
- 12:25because this sets up sort of like a
- 12:27cloud sandbox environment. Now, I'll be
- 12:29honest, I hardly ever do this because
- 12:31when I'm working locally, as long as I
- 12:33keep my machine on, as long as I keep my
- 12:35PC on, it feels pretty cloudlike. And
- 12:38here's what I mean by that. Right here,
- 12:39you can see that I'm mirroring my phone
- 12:40and I'm in the Chat GBT app. And if I go
- 12:42over here and I click on this right
- 12:44here, which says remote, this lets me
- 12:46see all of my actual codeex chats that
- 12:50are right here. So if I go to Herk 2 and
- 12:52I go to summarize past month, this is
- 12:54exactly what we were just looking at
- 12:55right here. You can see this is
- 12:56literally the same conversation. It has
- 12:58those three chats and I could come in
- 13:00here and I could say hi. And this gets
- 13:02sent in my actual codeex right here on
- 13:03my laptop or sorry on my desktop as well
- 13:06as my phone. So, as long as my machine's
- 13:08on, even though this is running locally
- 13:09and I'm out on a walk, I'm out at
- 13:11dinner, I could even be in a different
- 13:12state and I could control my desktop
- 13:15from my phone right here. So, let's say
- 13:17for some reason I left a file locally on
- 13:19my computer, I could just say, "Hey, can
- 13:20you email that file to me?" And then I
- 13:22could use it on my laptop. So, anyways,
- 13:23just wanted to show off that remote
- 13:25capability. And that's going to bring us
- 13:26into concept number six, which is work
- 13:29trees. Now, honestly, this is something
- 13:31where if you're not really building
- 13:32software or, you know, production apps,
- 13:34this isn't going to matter to you too
- 13:36much. Like I don't hardly ever use work
- 13:38trees, but it is good to know about
- 13:40since you will see these in the
- 13:41interface. So right here you can see
- 13:42that if I was to go to um the local
- 13:45setting, it says that you can work in a
- 13:47new local work tree which says create a
- 13:49copy of Herk 2 to work in parallel. So
- 13:52basically because I'm working in Herk 2,
- 13:54if I was going to do something like
- 13:55risky or if I wanted to try and
- 13:56experiment with something or for
- 13:58example, if I was building software and
- 13:59I wanted to try to change the way that
- 14:01the onboarding flow worked, but I wanted
- 14:02to like not interrupt the main branch,
- 14:05especially if other people on my team
- 14:06are working on that main branch, I could
- 14:08use a work tree to basically create a
- 14:09copy and I could test in there. And then
- 14:12what you can do later is you can merge
- 14:13it back into the main. So it's basically
- 14:15just a duplicate that lets you test. But
- 14:17like I said, there's hardly any times
- 14:18for knowledge work that I ever use work
- 14:20trees. And if I ever am building
- 14:21software and stuff, a lot of times Codex
- 14:23will say, "Hey, by the way, you might
- 14:24want to use this. You might want to do
- 14:25this in a new work tree." So, I wanted
- 14:26to call it out because it's important
- 14:28because you might see things like this
- 14:29master branch and you might see other
- 14:30work trees that might have been spun up
- 14:32automatically. But, I will say it's not
- 14:33something that I'm actively thinking
- 14:34about or think that you need to be
- 14:36actively thinking about in order to get
- 14:37the most out of codeex. So, wanted to
- 14:39bring that up. But, let's move on to
- 14:40number seven, which iscodeex.
- 14:43Now, this is really interesting. We
- 14:45talked about how when you're inside of
- 14:46your project, you have things like an
- 14:47agents.mmd. Now, if I go to my files,
- 14:50there's also something in here called a
- 14:52codeex. So, if I open this up, what
- 14:54you'll see in here is I have some agents
- 14:56and I have some workflows, but I also
- 14:58have a config. ML file, which is
- 15:01basically like some local settings. And
- 15:03there are basically going to be two
- 15:04different places where a codex file
- 15:06lives. And that's either going to be
- 15:08inside of your project. So like in this
- 15:10example, this is my codeex inside of my
- 15:12Herk 2. But there's also going to be one
- 15:14which is user or global, which means
- 15:16this lives under your actual user. And
- 15:18these are settings that hold things like
- 15:21memories and sessions and automations.
- 15:23These are things that will apply to your
- 15:25global codeex. So whenever you're in
- 15:27codeex, regardless of what your project
- 15:28you're actually in, the docex will have
- 15:31personal settings and app data and stuff
- 15:32like that. Now once again, I think this
- 15:34is something it's important to
- 15:35understand where that lives and what it
- 15:37does. But this isn't a folder that I'm
- 15:38actively thinking about and actively
- 15:40having to like maintain. But I did want
- 15:42to call that out because sometimes if
- 15:44you need to have a certain setting
- 15:45applied or something like that, that is
- 15:47where it lives. Now, the cool thing
- 15:48about this is if you're ever confused,
- 15:50you say, "Hey Codeex, what did you do
- 15:52there?" Or, "Do we need to add this to
- 15:53mycodex or do we need to move this to a
- 15:55different folder?" You can just ask it
- 15:57questions because it can search through
- 15:58its own documentation and understand how
- 16:00to do things like correctly. But I do
- 16:02think that understanding the difference
- 16:03between project scope things and user
- 16:05scope things that's definitely important
- 16:07especially as we get into the agents
- 16:09which is going to come up later. And
- 16:10that is a great segue into part three of
- 16:12the core concepts which is control and
- 16:14customization. So number eight here is
- 16:16about different AI models. So when you
- 16:20are starting a new chat you can choose
- 16:22the model you want to use. You can also
- 16:24change the model in between messages.
- 16:26So, I could shoot off a message on Astra
- 16:27and then I could, you know, change this
- 16:29to something else and then I could shoot
- 16:30off another message and I can keep
- 16:31changing if I want. Now, if you're
- 16:33watching this video way later, maybe the
- 16:34model names look different, but
- 16:35essentially the theory is the same. You
- 16:37have different models to choose from and
- 16:39they all have different prices and
- 16:41strengths. So, like right now, GBD6
- 16:42Astra is just the absolute strongest,
- 16:44but it's going to cost the most. It's
- 16:46going to eat the most of your weekly
- 16:47limit. Right here, you can see that I'm
- 16:48at 100% because it just got reset. But
- 16:51as you chat with Astra, it will drain
- 16:52your weekly usage faster than you know
- 16:555.6 Soul would, which would drain your
- 16:57usage limit faster than 5.6 Tarot would.
- 17:00Now, right here, I just want you to look
- 17:01at the API pricing so I can explain to
- 17:03you how these models are actually
- 17:05billing you. So, they build by tokens,
- 17:07and they usually build by 1 million
- 17:09tokens. Meaning, on Astra, for every 1
- 17:11million input tokens, it will cost you
- 17:14$10. And for every 1 million output
- 17:16tokens, it will cost you 50 bucks. And
- 17:17here you can see the different pricing
- 17:19for these different models. And you can
- 17:20see how it gets cheaper as you go down.
- 17:22Now, this is API pricing, which means if
- 17:24you were building an automation that was
- 17:26programmatically using these models,
- 17:28that's how much it would cost you. But
- 17:30this usage inside of Codex, that's not
- 17:31programmatic. This is usage limit or
- 17:34subscription limit. So, when you pay 20
- 17:35or 100 or 200 bucks a month for Codeex,
- 17:38this is what you're actually your usage
- 17:39limit is eating up is that subscription.
- 17:42And you're getting way more inference.
- 17:43Like, this is a $200 a month
- 17:44subscription. And if I use all of my
- 17:47weekly usage every single week, I'm
- 17:50getting about $14,000 worth of inference
- 17:53every month and I'm only paying 200
- 17:54bucks for that. So the subscriptions are
- 17:56much cheaper than if you were using the
- 17:57API pricing. But if you ever do go over
- 17:59your weekly limit, you'd have to buy
- 18:01extra codeex credits. I don't exactly
- 18:03know why they price it like this, but
- 18:04you know, you can see they just give you
- 18:06codeex credits instead. So let's say I
- 18:08went over my weekly limit. I could go
- 18:10here and I could click on buy credits.
- 18:12This would bring up this screen which
- 18:13lets you buy credits by like the
- 18:15thousands or something like that. And
- 18:16that is what I wanted to show you guys.
- 18:18So like 2500 credits, 5,000 credits,
- 18:2125,000 credits. And this is how it would
- 18:23charge you based on those tokens. Now,
- 18:25if you've never heard of tokens before,
- 18:26I know that might be a little confusing.
- 18:28Think about it as roughly four
- 18:29characters or roughly 3/4s of a word.
- 18:32And there's an OpenAI tokenizer right
- 18:33here where if I paste in some text, it
- 18:35will show you how many tokens this would
- 18:37be. So this paragraph for example was
- 18:40about 240 characters or sorry 274
- 18:42characters and about 56 tokens. And it's
- 18:44interesting because sometimes
- 18:45punctuation is a token, right? Like this
- 18:47period is a token, this comma is a
- 18:49token, this word's a token, um this
- 18:51word's a token. And this is basically
- 18:53just kind of breaking down what it might
- 18:54look like for these tokens. But anyways,
- 18:56you can see this was 56. And what else
- 18:58you might notice is that I earlier said
- 19:00input tokens or output tokens. Input
- 19:02tokens are whatever you're feeding in.
- 19:04So your prompts or if it's reading like
- 19:05a PDF or a word doc those are input
- 19:08because they're going into the model and
- 19:09output tokens are anything that it
- 19:11outputs. So even its reasoning here is
- 19:13output tokens. This entire thread is
- 19:15output tokens and that's why the output
- 19:17tokens are more expensive. Okay. So
- 19:19let's move on to number nine which is
- 19:20about effort. So within each model not
- 19:23only do you choose the model but you can
- 19:24also choose the effort level. So right
- 19:26here I can move Astra down to medium or
- 19:28I can move it down to low or I can move
- 19:30it up to extra high or if I go like
- 19:32ultra I have to allow full access
- 19:34because that can get you know pretty
- 19:36autonomous. But anyways the point being
- 19:38as you move up on these effort levels
- 19:40you get more quality like you get more
- 19:43intelligence but you also get more cost.
- 19:46It's more expensive to run it at higher
- 19:47effort levels. And that's why when you
- 19:49see these benchmarks, when there are new
- 19:50models released, you'll see the model,
- 19:52like for example, Cloud Fable 5, you can
- 19:54see this mark was on low and this one
- 19:57was on medium and this one was on high.
- 19:59So, it shows how one model can behave
- 20:01differently and also cost a different
- 20:02amount based on the effort level that it
- 20:04was being used and tested at. Now, it
- 20:06actually is pretty nice to play with
- 20:08these because sometimes Astra High is
- 20:10just so much more powerful than what you
- 20:11need to do. Like if you were using GBD6
- 20:13Astra High to help you write an email,
- 20:15that's overkill and you're paying way
- 20:17more than you should realistically to
- 20:18write an email. I could probably come
- 20:19down here and use 56 Terra and be just
- 20:21fine with that and it's a lot cheaper.
- 20:23So definitely try to play with the model
- 20:24and play with the effort based on your
- 20:26task. But a lot of times if you're just
- 20:27doing a lot of knowledge work, I would
- 20:28just chuck it on sole and I'd maybe just
- 20:30go to medium and just call it a day
- 20:32there. Now the other thing about effort
- 20:33is you can do something called fast
- 20:34mode. This obviously makes the model
- 20:36respond quicker and do things quicker.
- 20:37And it it's 1.5 speed, but it costs you
- 20:39more usage, so it will eat your weekly
- 20:41subscription faster as well. All right,
- 20:43and moving on to number 10, we have
- 20:45permissions. Basically, right down here,
- 20:47you can see that this is orange and it
- 20:48says full access. This is basically
- 20:50Codeex's yolo mode, which means that
- 20:52it's not going to stop and ask you a
- 20:54bunch of permissions and questions. Can
- 20:55I do this? Can I do this? If you go on
- 20:57ask for approval, it's going to always
- 20:59ask to edit external files and always
- 21:01ask to use the internet. So that's if
- 21:02you want to sit here and really sort of
- 21:04like manage it and you're wanting to
- 21:05make sure everything's safe, you know,
- 21:07maybe start off like this so you can
- 21:08just understand, oh, okay, it's asking
- 21:09me if it can run this command or it's
- 21:11asking me if it can open up my browser
- 21:12and then you just get familiar with what
- 21:14it looks like for Codeex to actually run
- 21:15that agentic loop. You could also go on
- 21:17approveof for me. So it's only going to
- 21:19ask you if actions are detected as
- 21:21unsafe like maybe deletes or maybe
- 21:22certain sorts of like API calls or
- 21:25things like that. So that's how these
- 21:26permissions work. And if you really want
- 21:27to get granular about, oh, I want to
- 21:29like make sure these certain actions
- 21:30never happen, you can ask Codex about
- 21:32that and it can work things into those
- 21:34config files like we talked about
- 21:35earlier with the codeex that it can
- 21:37certainly like block things out. But now
- 21:38let's move on to number 11, which I
- 21:40think is my favorite one, which is
- 21:42skills. Skills are so so important.
- 21:44They're basically reusable workflows.
- 21:46They're recipes that let you do
- 21:48something once and then teach Codex how
- 21:50to do it the same way every single time.
- 21:52So now you can just run all these
- 21:53skills. So like I said, a skill is
- 21:55basically just a recipe. So, let's say
- 21:57you come in here and you make, you know,
- 21:59a chocolate chip pancake and you
- 22:00followed a recipe to make that pancake.
- 22:02You would basically say, "Okay, if I
- 22:04ever want to make chocolate chip
- 22:04pancakes again, I'm going to use this
- 22:06recipe because I know that I'm going to
- 22:07get a good result." But let's say that
- 22:09you burnt these pancakes and you said,
- 22:11"Okay, well, what I need to do is make a
- 22:12quick update in this recipe to say, hey,
- 22:14you know, cook it up 30 seconds less on
- 22:16each side." And the next time you run
- 22:17the skill, you basically see, okay, are
- 22:19these better? Are these good? Do I need
- 22:20to make any more feedback in here? Or
- 22:22now, is this skill just the way I like
- 22:24it? And then your agents are able to use
- 22:26these skills. So here's an example where
- 22:28I basically gave it a YouTube video. So
- 22:29you can see that this is literally just
- 22:30a local file path of a YouTube video.
- 22:32And I said that I needed a video
- 22:34description, timestamps, and a LinkedIn
- 22:35post. It went ahead and it ran its
- 22:37agentic loop and it gave me a
- 22:39description, timestamps, and a LinkedIn
- 22:40post. And then I said, "Okay, now I need
- 22:42you to turn this into an X article." And
- 22:43put the draft into my X as a draft. And
- 22:45it knew exactly how to do this with a
- 22:47thumbnail, with a title. It also put
- 22:49screenshots from my actual video and it
- 22:50sprinkled them throughout my whole X
- 22:52article. And this is because it used a
- 22:54bunch of skills to be able to know how
- 22:56to do this. So right here, if I asked
- 22:57it, what skills did you use? It used my
- 22:59YouTube description skill, my timestamp
- 23:01skill, my LinkedIn post skill, my ex
- 23:02article skill, my ex article from video
- 23:04skill, my format X article skill. It
- 23:06used all of these skills because I've
- 23:07taught it how to do something good once.
- 23:09And then every time I run these skills,
- 23:11I just give feedback. Hey, that was
- 23:12good. You know, I liked this, but I
- 23:14didn't like this. Update the skill. And
- 23:15now what's cool is you can just invoke
- 23:17these either with natural language. It
- 23:18will be smart enough to do it with
- 23:19natural language, or you can do it as a
- 23:21slash command. So, I have one in here
- 23:22that I can call like the audit. And you
- 23:24can see this will run my audit or I have
- 23:26one in here called the 3D brain. And
- 23:28this runs the 3D brain. Man, I think I
- 23:30planned this out so good. This is going
- 23:31to bring us on to number 12, which is
- 23:34agents. So, we talked about codecs and
- 23:36agents is basically the exact same. It
- 23:39just holds different things. So, right
- 23:40here, you can see that the agents holds
- 23:42your skills and these once again can
- 23:44either be project level skills or they
- 23:45can be user scope skills. So, they can
- 23:47be global skills. So if I come into this
- 23:50chat right here which is inside of my um
- 23:53you know Herk 2 project and I go to open
- 23:55up the files and I go to my agents
- 23:58folder which is right here we have a
- 23:59folder called skills and this is a bunch
- 24:01of skills as you can see. Now if I open
- 24:03up an actual skill so for example let me
- 24:05go to my YouTube description skill. This
- 24:08is an actual skill.md. So once again MD
- 24:10just means markdown file. So what we
- 24:12have here is some metadata and this is
- 24:14what codeex will read to understand okay
- 24:16do I need to invoke the skill or not.
- 24:17So, it has the name and it has basically
- 24:18when to use the skill. And then all the
- 24:20skill is is just instructions. It's,
- 24:22hey, for Nate's YouTube descriptions,
- 24:25keep them short, two to four sentences,
- 24:26paragraph form, no bullet points, no m
- 24:28dashes. Here's the process. You read the
- 24:30video outline. You identify the core
- 24:32topic. You write a short paragraph. And
- 24:33that's the skill. It's a very, very
- 24:35simple prompt. But if I go to my ex
- 24:36article one, they can also be way more
- 24:38in-depth. So, this one is way longer.
- 24:40There's a lot of other rules and a lot
- 24:42of other criteria. There's way more
- 24:43steps. There's way more things to do.
- 24:45And also what's cool is skills can
- 24:47reference other things. Skills can
- 24:48reference other skills. They can
- 24:50reference other agents. They can
- 24:51reference Python files or context files.
- 24:54So in this example, this skill
- 24:55references my voice guidelines like my
- 24:57LinkedIn style guide as well as my X
- 24:58article style guide. But that is where
- 25:00your skills live inside of the agents
- 25:02folder. And then moving on to number 13.
- 25:04Here we have plugins. And this is where
- 25:06a lot of the magic actually happens
- 25:08because you can super easily come over
- 25:10here in the Codex desktop app. You can
- 25:12go to plugins and you can connect a
- 25:13bunch of things. You can see all the
- 25:14stuff that I have installed like Alpaca,
- 25:16Gmail, ClickUp, Clay, Google Drive,
- 25:18GitHub, Canva, and the list goes on and
- 25:21on. You can also search for different
- 25:23plugins. So, let's say you were really
- 25:24interested in connecting something like,
- 25:26I don't know, let's see, do they have a
- 25:27LinkedIn plugin? They have a LinkedIn
- 25:29plugin where we can find the right
- 25:30profession. We can grow our business
- 25:31with ads. We have all these other sorts
- 25:33of plugins, too. So, if you want to try
- 25:35to connect your different apps, just
- 25:36come in here and see if you can connect
- 25:38them using these plugins. Maybe they
- 25:39have Composio, they don't have
- 25:41Composeio, and if they don't have a
- 25:42plugin here, you can still connect it
- 25:44using AENV, using an API key. But
- 25:46plugins are so much easier because you
- 25:48can just basically sign in once. They've
- 25:50got ones for creativity, for developer
- 25:51tools, for business and operations, for
- 25:53data analytics, communication, and
- 25:55they're basically always growing this
- 25:57plug-in library. All right, so that now
- 25:59brings us on to part four, which is
- 26:01tools and scale. Okay, we've got five
- 26:03left here. Number 14, we have the
- 26:05browser. Now, this is definitely one of
- 26:07my favorite things about Codeex. If I
- 26:10open up the tab over here, we can open
- 26:11up a browser, which is really cool
- 26:13because I can basically just control
- 26:14this. So, like I could go to school.com.
- 26:16And what you'll notice is when I log
- 26:18into school.com over here, I'm just
- 26:20going to zoom out a little bit. You can
- 26:21see right here that I'm already logged
- 26:22into my school account. So, I could
- 26:23manage my different communities, which
- 26:25also means that Codeex can help me
- 26:27manage my communities from here. Because
- 26:28when you save these login inside of the
- 26:30Codeex browser, it then saves them. So,
- 26:32if you say, "Hey, could you go to that
- 26:33account and pull the report for me?" or
- 26:35hey could you go into school and could
- 26:36you make this post for me it can do it
- 26:38through the browser if there's not a
- 26:40plugin or if there's not like an API or
- 26:42MCP server and the browser use of codeex
- 26:44and specifically with GBT6 Astra is the
- 26:47best that I've ever used ever I'll play
- 26:49a clip right here where I asked it to
- 26:51basically open up Canva and paint me and
- 26:53draw me using the tools in there and I
- 26:55gave it just a picture of me and it was
- 26:56able to replicate it in a way that I
- 26:58thought was really really good. So silly
- 27:00example, but it definitely demonstrates
- 27:01the vision and the browser use
- 27:02capability of this agent. All right, so
- 27:05number 15 we have sites. So if you go
- 27:07over here to explore and you click on
- 27:08sites, this is very very cool because it
- 27:11basically lets you put things out there
- 27:13onto the cloud. Remember earlier we
- 27:15talked about local versus cloud. Codex
- 27:17can build something for you and it can
- 27:18be a local host but then you say, "Hey,
- 27:19can you just put that on a site real
- 27:20quick so that you could have your team
- 27:22log in or so that you could share it
- 27:23with, you know, other people and make it
- 27:25publicly accessible." Like right here,
- 27:26you can see this London and Paris one is
- 27:28everyone whoever has this URL could go
- 27:30ahead and open this up. Whereas this one
- 27:32is just me. So it would literally have
- 27:34to be logged into my account in order to
- 27:36be able to open up this site. And what
- 27:38else is amazing about that is you can
- 27:39open up the analytics for the site. So
- 27:41you can basically host things. You can
- 27:43connect your own domain and this is
- 27:45basically going to replace something
- 27:46like a Verscell or like a I don't know a
- 27:49Squarespace, wherever you wanted to host
- 27:50your sites normally. You can now just do
- 27:51it right here from the Codex desktop app
- 27:53which is pretty cool. You can also even
- 27:54connect a database. So on the back end
- 27:56if you had to store like user login data
- 27:58or you know permissions or settings or
- 28:01you know a database of customer records
- 28:03whatever it is you could actually store
- 28:05that here too and it's really cool. All
- 28:06you need to do to actually use a site is
- 28:08you just basically have to do either say
- 28:10hey can you turn that into a site or you
- 28:11could do slash site and then it will
- 28:13basically know that you can you know
- 28:14host a site or build a site. So great
- 28:16touch here from codeex. All right so
- 28:18number 16 we have sub aents. So, sub
- 28:21agents are really cool because it lets
- 28:23you delegate work out to a bunch of
- 28:25different CEX agents. So, right here in
- 28:26this main chat, you can see that I'm
- 28:28talking to GBT6 Astra, which is a
- 28:30really, really smart and intelligent,
- 28:32but also expensive model. So, let's say
- 28:34I wanted to do a bunch of research, but
- 28:36I didn't want to waste Astra's brain on
- 28:37that. Or let's say I wanted to do like a
- 28:39bunch of different testing and just do a
- 28:40lot of stuff at once. Astra here that
- 28:42I'm talking to can delegate workout to
- 28:44tons of sub aents. Hey, I need you to
- 28:46search through X and search through
- 28:48YouTube and also dig through my own
- 28:49comments. So, just find, you know, what
- 28:50are people talking about right now in
- 28:52the AI space? Is there any big news or
- 28:54drama? I don't want you to do this
- 28:55research though. I want you to basically
- 28:57just delegate a bunch of sub agents out
- 28:58to do the work. And I want all of these
- 29:00sub aents to be using 5.6 Terra as the
- 29:03model. So, shoot off those agents and
- 29:05let me know what they find. So, this is
- 29:06really cool because I can specify that I
- 29:08want those sub aents, those little like
- 29:10researchers and workers to use different
- 29:13models. So right here it's using its
- 29:14agentic loop and it's going to spin up
- 29:16these agents. And whenever you see these
- 29:18little colorful things, these are
- 29:20different agents that it's spun up. So
- 29:21right here, this one's called X. And
- 29:23when I open up X, it is actually this
- 29:25sub agent working right here. I can
- 29:27watch this sub agent go through its
- 29:29agentic loop and I can watch it like
- 29:31pull in data and everything like that.
- 29:32So this is the same thing as over here,
- 29:34but this one is just a different model
- 29:36and it's not the main session that
- 29:37you're currently talking to. You can see
- 29:38we have two others that started working.
- 29:40We have this one called YouTube Pulse.
- 29:41And once again, it's going through the
- 29:42same exact aentic loop. We have this one
- 29:44called X drama. We have this one called
- 29:46audience comments. And we can basically
- 29:48watch all of these work, but we don't
- 29:50have to manage them because once these
- 29:52agents are done, the main session here
- 29:54will say, "Okay, cool. All five are
- 29:55done. They reported back to me. Here's
- 29:57what they found." And this is really
- 29:58good for gathering different
- 29:59perspectives or researching out a bunch
- 30:02of things in parallel. And these
- 30:03intelligent models are really, really
- 30:05good at delegation. So sub agents are
- 30:07definitely something that you need to be
- 30:08using. So, while these sub aents are
- 30:10fanned out and they're working, let's
- 30:11move on to number 17, which is scheduled
- 30:14tasks. Scheduled tasks are awesome
- 30:16because this basically lets you have
- 30:18your codecs working autonomously for
- 30:19you. So, on the lefth hand side, if you
- 30:21click on scheduled, you can basically
- 30:22ask codecs to schedule things for you,
- 30:25reminders, updates, or run full sessions
- 30:28because all a task basically is is it
- 30:30injects a prompt into a codec session
- 30:32like this. So it uses your same local
- 30:34files, it uses your same skills, it uses
- 30:36your model, it uses everything, the
- 30:38agentic loop, it just does it
- 30:39automatically for you. So in here, you
- 30:41see that I could create one and I could
- 30:43set it up manually. I could have it run
- 30:45locally or I have could have it run on
- 30:46the cloud. I could also choose what
- 30:48project to work in. So Herk 2 or, you
- 30:50know, Hyperframes editor. I can also
- 30:52choose what chat. So every one of these
- 30:54routines could be a new chat or it could
- 30:56be a new chat for this task or it could
- 30:58be even in an existing thread that we
- 31:00already have going. And then you can
- 31:01basically choose when to schedule it. So
- 31:03it could be daily, it could be hourly,
- 31:05it could be, you know, weekly, or you
- 31:07could get it really, really custom in
- 31:08here. And the cool thing is you can
- 31:09describe everything in here manually, or
- 31:11you could basically just ask Codex to
- 31:14make one for you. So, for example, I
- 31:15have these like seven or eight different
- 31:18ones that are running for um trading.
- 31:20They're literally trading $10,000 of my
- 31:22real money. And I have this set up as
- 31:24scheduled task so that they can
- 31:25basically all work together, talk to
- 31:26each other, check the market, make
- 31:28trades, things like that. And of course,
- 31:29you guessed it, these run in a different
- 31:31project right here called trading
- 31:33challenge. And they all run inside of
- 31:35this challenge thread. You can tell that
- 31:36this has scheduled tasks in it because
- 31:38there's a little clock icon next to that
- 31:40thread. And moving on to the very last
- 31:42concept we have for today, number 18,
- 31:44this is voice mode. Now, everything that
- 31:46you've learned so far, skills, sub
- 31:49aents, projects, agents, all of this can
- 31:51be controlled by using your voice to
- 31:53have a real conversation with codecs.
- 31:55With your voice, you can manage
- 31:57different codeex threads. You can
- 31:58delegate different agents, you can run
- 32:00skills, you can set up schedule tasks,
- 32:01you can build sites, you can do anything
- 32:03here with your voice. And all you have
- 32:04to do is come down to the bottom left
- 32:06and click on start new voice chat. So,
- 32:08forgive me here as I'm awkwardly like
- 32:09hunching down because my headphones
- 32:11don't reach far enough. But watch this.
- 32:13Hey, Codex. So, I need you to help me
- 32:15out real quick inside of my Herk 2
- 32:16project. Can you start a new thread to
- 32:18make a thumbnail? I'm making this
- 32:20YouTube video about 18 CEX concepts. So,
- 32:22if you could make me a thumbnail for my
- 32:24YouTube channel where you see me sitting
- 32:26at a desk working on a laptop and I'm
- 32:28pointing to the right and there's like
- 32:3018 different little AI agents. Um, yeah,
- 32:33just start start with that. And then
- 32:34once you have finished that thumbnail,
- 32:37could you send it to the thread that is
- 32:39currently working inside of my HERK 2 as
- 32:41well and it is doing research on YouTube
- 32:43and X and um looking at my channel and
- 32:46stuff like that. Could you just send
- 32:47that thumbnail to that thread?
- 32:48>> All right, let me check on that. Okay,
- 32:50I've got your thumbnail brief queued up
- 32:52with you with the laptop pointing toward
- 32:53the 18 ages using your existing photos
- 32:56for reference. And now found your
- 32:58research thread. So I'll send the
- 32:59finished thumbnail there. Now, do you
- 33:00guys see how this is all coming
- 33:01together? It started a work tree chat.
- 33:04It was doing the agentic loop. It also
- 33:06created a goal down here. It literally
- 33:07created its own goal because it knew
- 33:10that we had this, you know, task for it
- 33:12and it started working. And you know,
- 33:13the voice chat's still going. So, I
- 33:15don't know what it's going to think of
- 33:16this, but I just wanted to call that out
- 33:17because all of what we just talked about
- 33:19is all coming together because even you
- 33:20can see over here, it started making
- 33:22that actual prompt. Like, it shot off
- 33:25this prompt right here and it started
- 33:26off this new thread. And now, this is
- 33:29working in a work tree and it's going to
- 33:30create that thumbnail for us and then
- 33:32it's going to send it over into this
- 33:34actual chat and we'll see that actually
- 33:35happen. But as you keep navigating, like
- 33:37you can go to the web and you can go to,
- 33:40you know, other things and you can keep
- 33:41this voice working. So as you're just
- 33:43working on your laptop, it'll show you
- 33:45what's going on and you can keep talking
- 33:46to it. And you can also do all this on
- 33:48your phone as well because of the whole
- 33:49remote thing, you can use voice mode
- 33:51when you're on your phone, too. Okay, so
- 33:53that finished up and you can see what
- 33:54happens is it actually can send tasks or
- 33:58messages to each other, right? So, what
- 34:00happened was this was the original
- 34:01thread that it created to make the image
- 34:03and it made this and then it sent it to
- 34:05this thread and said, "Hey, Nate wanted
- 34:06me to send this to you. Here it is." So,
- 34:08with just your voice, I know that was a
- 34:10very simple example, but you're able to
- 34:12coordinate a bunch of different codec
- 34:14threads, manage them, multitask while
- 34:16you're on the go, while you're flipping
- 34:17between different apps. Very cool. But
- 34:19anyways, guys, that is going to do it. I
- 34:20hope you enjoyed. I hope you learned
- 34:21something new. But if you want to keep
- 34:23diving into more stuff with Codex, and
- 34:24I've made so many videos, I'll tag a
- 34:26playlist right up here that has just a
- 34:28bunch of Codeex content. So, hope to see
- 34:29you guys over there. Thanks for sticking
- 34:30to the end, and I'll see you in the next
- 34:32one. Thanks everyone.
About this transcript
This page contains the full transcript of Master 96% of Codex in under 35 minutes by Nate Herk | AI Automation, generated from the public captions YouTube serves with the video. The transcript has 8,664 words across 1,175 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.
What you can do with it
Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to choose a caption track available on the video.
Free YouTube transcript tool
YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, caption language selection, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.