STOP Using Claude Code Without This Fable 5 Agentic OS
CChase AI
Computing/SoftwareSmall Business/StartupsInternet Technology
Transcript
00:00:00Hey Jarvis, give me the rundown for today.
00:00:03Good morning.
00:00:04You're at about 466,000 followers across all platforms, up about 3,000 this week.
00:00:12The latest videos pulling about 4,000 views a day, 17,000 so far.
00:00:18Big story in AI today, U.S. government directive forced Anthropic to suspend public access to its top mythos class.
00:00:27Biggest thing on today's board, cut and ship the Jarvis HUD reveal video.
00:00:33Want me to run the daily inbox audit or do you have anything else in mind?
00:00:37So what exactly are we looking at here?
00:00:39Well, this is Jarvis, our Claude Fable 5 OS.
00:00:43Now, when we say Fable 5, I mean this was built using Fable 5, but it does not require us to have Fable 5 to actually run this.
00:00:51In fact, a lot of what you're seeing here is actually completely local, and it can run on essentially any model you want.
00:00:58Now, if you have seen my previous Agentic OS videos, then you kind of know the deal here.
00:01:01We've taken Cloud Code and we've added another layer on top of it, which gives us some things that you just can't get inside the terminal.
00:01:08Things like visual metrics, things like skills that are turned into buttons, and this isn't a productivity theater thing.
00:01:15This is something that gives us a true boost if we're someone who's operating in a bunch of different domains, and is also something we can easily package for clients or members of our team who aren't particularly technical.
00:01:26And Jarvis is just the next evolution of this Agentic OS model.
00:01:30Its backbone is still this robust, completely customizable, Claude Code skill architecture that takes everything you do in your day-to-day, your manual workflows, your daily tasks, and turns those into skills and automations.
00:01:43And it's on top of that bedrock that we build this.
00:01:46And in today's video, I'm going to show you how it works, where the true value lies, and how you can create something like this for yourself.
00:01:53And I think there's a lot of things you can take from this project, especially the local voice model dynamic we have going on.
00:01:58Before we dive into all that, a quick word from today's sponsor, me.
00:02:02So inside of Chase AI+, you not only have access to my exact setup you see in today's video, you also get the Claude Code masterclass, which is the number one way to go from zero to AI dev, especially if you don't come from a technical background.
00:02:14I update this every single week, and it also includes a Codex masterclass.
00:02:18So if you're someone who's trying to stay on the cutting edge of AI, this is the place for you.
00:02:23We're currently running some deals on the membership, so if you want to take a look, just check out the pinned comment.
00:02:28So let's begin with a quick lay of the land with Jarvis so you can understand what it is you're actually looking at here.
00:02:34After we do that, we'll take a look under the hood, see how this is actually working, so you understand how to customize it and how to build it yourself.
00:02:41So front and center, we have the whole voice aspect, again, completely local, which allows it to be relatively fast and snappy compared to routing this all through something like 11Labs, for example.
00:02:53And in the beginning of this video, you heard Jarvis give me a whole spiel about what's going on with my latest videos, what's going on with my follower account, AI news, that sort of thing.
00:03:01That is not a hard-coded script.
00:03:03What happens is when I ask Jarvis, hey, give me the rundown for today or what's happening today, it takes a look at the various reports that are automatically generated inside my Obsidian Vault and determines what's actually important out of those reports and what I should know about.
00:03:18As it did that, you will remember, there were some various pop-ups, and these little pop-ups are reports or links to things that are relevant based on what it's talking to you about.
00:03:29So remember, it said like, hey, your last video is doing X, Y, and Z, so it has this latest deploy pop-up, which takes me to said video.
00:03:37It mentions stuff about Anthropic, you know, essentially getting Fable 5 canceled by the government.
00:03:42What does it do?
00:03:43It brings up the source article for that.
00:03:44It also talked about more things related to AI news and what's trending, and that all came from the morning report.
00:03:49So if I click this, you see this full write-up.
00:03:52This write-up lives inside Obsidian.
00:03:54This whole system is linked to Obsidian.
00:03:56So while I can read it right here, I can also click Open in Obsidian, and it brings up the original report inside of Obsidian.
00:04:04I can click on the different links, so there's a whole connection.
00:04:07You also remember it asked at the end of its spiel, hey, do you want me to do that inbox triage for you?
00:04:12Do you want me to go inside your Gmail and see what's important, what we need to respond to?
00:04:15Well, that is a skill, and those skills and automations are represented over here on the right.
00:04:21This allows me to instantly run any Cloud Code skill or automation with a click of a button.
00:04:26And again, if you remember what we did with our Cloud OS that we created instead of Obsidian the other week, same exact idea.
00:04:33The value add here is more for if you're using this with a non-technical team or non-technical client,
00:04:39and they want to be able to run skills and automations with Cloud Code, but they're not the type to ever open up the terminal.
00:04:44This allows them to get all that power with a click of a button and again, fully customize.
00:04:49So let's say I did want to get a full inbox brief.
00:04:52If I just click inbox brief over here on the top right, you can see it mentions that it's queued right away.
00:05:00We see it here, and then we also see a new pop-up showing inbox brief, and we get a little progress bar showing that it's working on said automation.
00:05:09And once it finishes that, it'll not only generate a written report, it can actually give me a verbal rundown of, hey, here's what's going on, here's what you should care about.
00:05:16So it gave me the quick verbal rundown of what it found, and then I can see the actual report, which again, as always, is linked inside Obsidian.
00:05:34So these pop-ups are useful, they're relevant, and they link us to things we actually care about, and at any time I can clear them out.
00:05:39Now let's look over here on the right a little bit more.
00:05:41So kind of already talked about it.
00:05:43We have these different skills that we can change out at any time.
00:05:45One click away, it runs them.
00:05:47Below that, we have the schedule.
00:05:50So this is just sort of my daily schedule that is linked to my Google Calendar.
00:05:53If I click it, it brings up my calendar.
00:05:56We have a little audio section.
00:05:58So if you're paying attention when it was actually talking to us, we could see it kind of just moving up and down, and it just lets you know if it's actually working.
00:06:05And below that, we have a summary of what's going on in terms of AI news.
00:06:10Now, everything you see here, along with everything we talked about, completely customizable.
00:06:15Maybe you don't care about having your Google Calendar schedule here.
00:06:18Maybe you don't care about showing the audio or having an AI wire of the news.
00:06:21That's fine.
00:06:22The beauty of these agentic OS, you know, sort of systems is the fact that it isn't a one-size-fits-all.
00:06:29This is just a set of tools that I'm showing you that you can take and do whatever you want with them.
00:06:34You know, your metrics will be different than my metrics.
00:06:36The sort of things you want to have one click away are going to be very different than what mine are.
00:06:40It all depends on sort of your daily workflows and what you or your business do.
00:06:44But the plus size of something like that is like when we talk about creating this for clients or even team members is like you can make it whatever you want it to be for them.
00:06:51Here on the left side, kind of the same thing.
00:06:53For my vitals, I show things like my subscriber counts, what's going on with my latest video.
00:06:58I have a little tracker for my Claude tokens over the last five hours.
00:07:01And some things like directives.
00:07:03So directives are just like, hey, here's like the top three things you should be working on today.
00:07:06Again, totally dynamic and Claude code figures that out based on my schedule.
00:07:10And then I have a little documents trail.
00:07:13So everything that's created or referenced by Jarvis doesn't just come up as a pop-up.
00:07:18It's over here.
00:07:18So if I want to click it, again, like the morning report, it's right there.
00:07:22So that's the user interface.
00:07:23That's the visual side of Jarvis.
00:07:25Now let's talk about the actual nuts and bolts.
00:07:28What's actually going on under the hood here?
00:07:29Because that's what actually matters, let's be honest.
00:07:32If we kind of just stopped here and it was a fancy UI, well, again, we're kind of just talking about productivity theater.
00:07:37If this is to be something that actually moves the needle versus sitting in the terminal all the time, it needs a proper backbone.
00:07:43And that's what we're looking at here.
00:07:45So let's kind of walk through what happens when you talk to Jarvis and ask it to do something because it can go down a number of different paths.
00:07:53So here you are, and let's say you tell Jarvis, give me an update on today's morning brief.
00:08:00You have some sort of automation that you run every morning.
00:08:03It grabs whatever information is relevant to you, and you want Jarvis to tell you about it.
00:08:07You don't want to read it.
00:08:07You want it to actually either run it or if it's already been run, tell you about it.
00:08:12So you audibly tell that to Jarvis, hey, what's going on in today's morning report?
00:08:18That voice, because again, you used your microphone for this, that audio goes to Faster Whisper.
00:08:25Now, Faster Whisper is a free, locally sourced program that is going to take what you spoke and it's going to transcribe it.
00:08:34Now, there are a million and one different, you know, local audio transcribers.
00:08:39You can replace this with whatever you want.
00:08:40This is just the one I chose because it works just fine.
00:08:42So it's taking your voice and it's turned that into a transcript.
00:08:46Now, what do we do with this?
00:08:47Because we've talked about Fable 5 a bit and all these things.
00:08:50Does every transcript then need to get sent to Cloud Code and we run Fable 5 and it takes a look at it?
00:08:55No, of course not.
00:08:56This whole thing is Fable 5 built, but most of the stuff that's running under the hood is either going to be Opus or Sonnet,
00:09:02or frankly, you could use a local model for a lot of these things because Jarvis isn't, while it can do it,
00:09:08you can tell it to use Fable and you use it just fine.
00:09:11It's not like you're creating projects out of this.
00:09:13This is more in the personal assistant realm or for a non-technical team that just wants it to do
00:09:17a lot of the similar things all the time with a nice, easy to understand UI.
00:09:22So just, that's kind of like big picture.
00:09:24But so we've taken what you've talked about, we've turned it into a transcript.
00:09:26Now we need to figure out what to do with it.
00:09:28And we kind of have three options.
00:09:30Now the first question that's going to be asked is, can we route this transcript or this question via regular expressions or regex?
00:09:39So what's going to happen is this essentially script, this deterministic piece of code, AI isn't involved here,
00:09:45it's going to look at your transcript and say, hmm, did it match some of these specific pre-coded phases
00:09:52that we know need to automatically do something?
00:09:55What do we mean by that?
00:09:56Well, in the intro, what did I say?
00:09:58I said the rundown for today, right?
00:10:00I use the term rundown.
00:10:03Rundown is a specific trigger word, so to speak, that regex picks up and automatically routes it to do a specific thing.
00:10:11And that specific thing is what you heard in the beginning.
00:10:14Take a look at the reports, tell me what's going on, give me a rundown for the day, right?
00:10:19A basic summary, so it automatically catches that keyword and does a specific thing.
00:10:24We like this because since it is just code, it doesn't cost us any money, AI isn't involved, and it's very, very fast, right?
00:10:32In a perfect world, we could do everything like that, and we have very, very quick responses.
00:10:36But that isn't everything because most of the time what you're telling the AI system to do can be somewhat murky, right?
00:10:43It requires a little bit of intelligence to figure out what to do and how to route it.
00:10:47And that's where option two comes in, and that is where we bring in Haiku.
00:10:51Now, why are we using Haiku?
00:10:52Why are we using the dumbest anthropic model out there?
00:10:54Well, because it's cheap and it's fast.
00:10:57What we're doing here in this whole section is we're simply routing.
00:11:01I'm not doing anything unless it's regular expressions that I know I can route it right away.
00:11:06We're just trying to figure out at this phase where does this response need to go to and who needs to do it.
00:11:11So Haiku is actually very good at this.
00:11:13It's the most of the things you're going to be asking Jarvis to do are not too complicated.
00:11:18It just needs to understand which path to go down, and Haiku figures that out for us.
00:11:22Again, we're talking about fractions of a penny for each request.
00:11:26Now, the third option, and one that I include in my system, but, you know, it kind of just depends on what you want to do, is to have this be a completely local model.
00:11:34You don't have to use Haiku at all.
00:11:35You can use a model on your local computer, which, again, leans into the local side of this whole system, and it does the same thing.
00:11:42It's going to take a look at the transcript to figure out where to route it.
00:11:45We're just routing here.
00:11:46So to recap, you talk, it gets transcribed, and then we either use regular expressions, Haiku, or a local model to figure out what to do with it.
00:11:54So zooming out here, we've now given Jarvis that request.
00:11:59Tell me what's going on in the morning report.
00:12:01So what's going to happen is Haiku is going to say, okay, let's take a look at Obsidian.
00:12:06Does this exist already?
00:12:08If this is a report that's already on the disk, well, that's something we can very quickly and easy take a look at, read, and respond to.
00:12:16If it does not exist inside Obsidian, well, what's it going to do?
00:12:19Well, it's then going to tell Claude Code to create the morning report.
00:12:24Now, again, Haiku is routing it, but Haiku is not going to be the one who executes this.
00:12:28We want a stronger model.
00:12:30So by default, it will be Opus, but you could change that to Sonnet.
00:12:34And if you're crazy, you could change that to Fable 5 when it eventually comes back.
00:12:38So if it hasn't been created yet, what are we going to do?
00:12:42We have a headless version of Claude Code.
00:12:44It's like opening up Claude Code, but it's invisible.
00:12:47And it uses dash P.
00:12:49Now, of note, using headless Claude Code with dash P in about a day or so is not going to pull from your subscription.
00:12:57Okay?
00:12:57It's going to pull from that $200, like additional API credits to give you every single month.
00:13:02Can that be a problem?
00:13:04At huge scale?
00:13:05Yes.
00:13:06Which is why you want to do a lot of these things with Sonnet.
00:13:10Is it in reality going to be a problem?
00:13:13I would argue not really.
00:13:15What is the purpose of something like Jarvis?
00:13:17It's to act in this personal assistant, like task manager.
00:13:21We aren't building Facebook with Jarvis.
00:13:24You know, if you're actually doing that, you would be hardcore in the terminal.
00:13:27This is personal assistant type things, things for non-technical team members.
00:13:31At a certain complexity, you wouldn't use this.
00:13:33This isn't the tool for the job.
00:13:35That being said, because of that, is it likely you would run through $200 worth of credits if you're using Sonnet for the most part?
00:13:44I would argue no.
00:13:46Not a big problem, but something to think about.
00:13:48And again, if that is an issue for you, then just don't use Cloud Code for this.
00:13:52This is called the Fable 5 OS, but let's be honest.
00:13:55You could replace that with Codex.
00:13:56You could replace this entire thing with local models.
00:13:59You know, it doesn't have to be Cloud Code.
00:14:02The infrastructure is customizable and can be swapped out with whatever you want.
00:14:07But back to our example, we're saying we wanted the morning report.
00:14:10It wasn't inside Obsidian, so now it's going to spawn a headless version of Cloud Code to create the report for us.
00:14:16That report then gets uploaded to Obsidian, it reads the report, and then it generates a response with that summary.
00:14:25That summary goes to Kokoro.
00:14:29Kokoro is another local open source model that is going to take the summary transcript that Cloud Code has created and turn that into words, turn that into a voice.
00:14:39Think of it as like a mini-11 labs on our computer.
00:14:43Then Kokoro is going to say, hey, in today's morning report, blah, blah, blah, blah, blah, and that goes up to you.
00:14:49The voice that you heard today with Jarvis is from Kokoro.
00:14:53That could be swapped out with any voice you want.
00:14:55So again, local, customizable.
00:14:56But that's how this whole thing runs in terms of you ask Jarvis to do something, and then that's where the route it goes.
00:15:04So to summarize one more time, as I'm sure many of you are very confused, what happens?
00:15:10You ask Jarvis to do something.
00:15:13That ask needs to be turned into a transcript.
00:15:16Faster Whisper does this for free.
00:15:18Once we have the transcript, we need to figure out, how do we route this?
00:15:22We either use regular expressions, a cheap haiku model, or a local model.
00:15:28From there, we usually need to figure out, is this something that already exists in Obsidian, or do we need to call on Cloud Code to generate this report, this request for us?
00:15:38Once it generates what it needs to generate, either it's going to create a report inside Obsidian, or maybe it's an HTML page.
00:15:44Either way, it's going to do what it has to do, it's going to send the response to Kokoro, which turns that into a voice we can hear, and that comes back to us.
00:15:53Simple enough.
00:15:54Now, going a step further, everything that's actually going on under the hood.
00:15:57When we say morning report, when we say do skill A, B, and C, what are we actually talking about?
00:16:01Well, we're talking about this.
00:16:03We're talking about the skill architecture that really is the backbone of everything.
00:16:08Because what is a morning report?
00:16:09Well, that could mean anything.
00:16:10What that is, that is a skill, is a skill that is made up of other miniature skills that give me a large report saying, go check these sources for this information, go check these social media pages to get this information, yada, yada, yada.
00:16:24But, like I keep harping on, this skill architecture is only as strong as you make it.
00:16:30And how do we make this?
00:16:31And if you've watched my other videos, you know my spiel on this.
00:16:34What you need to do is we need to figure out a way to take your daily workflows.
00:16:38When I say your daily workflows, what am I saying?
00:16:40I'm saying, what do you do day to day in your personal life or your business?
00:16:43Like, what are these common tasks that are repeated over and over and over?
00:16:47Have you, someone at this point who probably knows how to use Cloud Code, have you actually sat down with Cloud Code and said, hey, here's what I do every day.
00:16:55Can we break all that stuff out in individual tasks?
00:16:58And then can we turn those tasks into skills?
00:17:01And if it makes sense, turn those skills into automations.
00:17:04That's how you build a skill architecture like this.
00:17:06And this is where all the power of this lies.
00:17:08For example, and what you see here is a lot of what I do in my different skills, something like content research.
00:17:14Again, most of you aren't creating content, so it'll be something completely different.
00:17:17But, the point remains.
00:17:19So, what do I do for research?
00:17:20Well, obviously, I need to take a look at things like YouTube.
00:17:24And so, I turned searching on YouTube for information into something called my YT Pipeline skill, where it takes a look at YouTube videos, sends them to Notebook LM, and then gives me summaries.
00:17:35I need to be able to do deep research on certain topics that goes beyond YouTube, that takes a look at other stuff, whether that's Twitter, the internet at large, et cetera.
00:17:43Well, I turn that into an entire customized deep research skill.
00:17:47I have an entire GraphRag system with LightRag, which has a lot of information about what I've done in the past.
00:17:52So, what did I do?
00:17:53I turned that entire query system into a skill.
00:17:56On, and on, and on, and on.
00:17:58I then repeated that across all the different domains of my personal and business life, whether that's content, my community, my agency, sales, et cetera, et cetera, et cetera.
00:18:07And in practical terms, how you would do that is quite literally, you would just open up Cloud Code, and you give it a stream of consciousness saying, here's what I do every day.
00:18:15And then you'd be like, hey, can we turn those into skills?
00:18:17And inside of Chase Heia Plus and the link below, I have like a full script that you can plug in a Cloud Code, and it will walk you through that if you need it.
00:18:25And to sort of tie that all in above, when we talk about these skills, if you then tell Jarvis, hey, do skill X, Y, and Z, and it goes through the transcription process and the routing, the part where we bring up headless Cloud Code, you know, and it's running this in the background to do something.
00:18:42If what it's doing is just a simple skill, you know you're going to get an output that's A, correct, and B, consistent, because you've already mapped out the process.
00:18:53There's nothing really left up to chance.
00:18:55And any time we're able to create AI systems that are more deterministic in nature, the better.
00:19:02The less we are subject to just the issues of AI in general, like, well, maybe it will do what I want, maybe it won't, right?
00:19:08When we create skills, we've codified certain things.
00:19:11And so you take that all together, and you have Jarvis, you know, and you can see under the hood, it's a lot more than just like a fancy UI with some metrics that are nice, that are just nice to have.
00:19:22There's a lot going on underneath it.
00:19:23And again, the true power of something like this, because it is a web app, it's the fact that we can bundle it and share it with team members and share it with clients.
00:19:30They just have to tell you, if you're the one creating it for them, what they want on the customization side.
00:19:36And the real, real power is the whole skill architecture, and the fact that you're going to sort of walk them through the codifying of different tasks.
00:19:46Because then they can sit here, or anyone could sit in this chair I'm in right here, and get like 80, 90% of the power of Claude code in the manner that I use it every day, which is these different skills and automations.
00:19:56It's literally a click away from them.
00:19:58That's what you really give them.
00:20:00And then if they want something else, you know, we have the whole voice piece of it.
00:20:03But this, in the end, is a fancy layer on top of all this.
00:20:07So that's the whole system in the nutshell.
00:20:09I really like it, mainly because of the customization of it, and the fact that we can add a lot of local stuff.
00:20:14So you can get pretty creative with what it's able to do and what it's able to connect to.
00:20:18Again, nothing's stopping you from bringing in more outside sources of something like this, like turning it into like a Slack agent and that sort of thing.
00:20:26That's where I'm going to leave you with this, guys.
00:20:30If you want, again, my exact setup inside of Chase AM+, there's a link to that in the pinned comment.
00:20:36But otherwise, let me know what you thought, and I'll see you around.
Community Posts
No posts yet. Be the first to write about this video!
Write about this video