
Loading summary
A
I built this Claude OS system basically. This system is essentially like a collection of a bunch of skills, a bunch of little automations. Every time I add a new tool or a new MCP or CLI or something, I just say add this to your tools MD overnight. It does the streaming. It goes and checks daily memory files and then creates a compressed version of it inside of the long term memory file. It takes what I've ordered last week and just puts it into the cart again. So it's basically like helping me automate my grocery shopping. Just get started. The most important part is getting like this initial structure in place and then once you have that, just like starting to connect your tools one by one.
B
Hey everyone. So one of the biggest first world problems that we all have now is should we set up a personal chief of staff in Open Claw or cloud code? That's why I'm really excited to welcome my friend Moritz to talk about the pros and cons of each system. Then he'll dive into exactly how he set up cloud code to be his chief of staff, from his folder structure to skills. So welcome sir.
A
Thank you, Peter. Happy to be on.
B
Yeah, it's great to have you, man. So why don't we dive into it, I think a particular diagram that shows how OpenCloud and Cloudcall works. Can you share that?
A
Yeah. So I put this diagram here together. So basically comparing OpenClaw and Cloud Code and we can go into like the pros and cons of each because each have like certain pros and certain cons as well. So let's start with the openclaw side. So I would say one of the top pros of openclaw still is like this aspect of mobile access. It was built that you can bring it into a huge range of different chat applications like Telegram, Discord, Slack and so on. That's just, I think one of the standout features, also one of the things that made it so popular. And if we compare that to what like Claude Code has built over time, they've built this dispatch at some point and then they've also recently built like the Telegram and Discord plugins which work fine I guess, but it's still not as good as the like actual experience of using OpenClaw through, through these chat applications. Would you agree with that?
B
Yeah, I agree with that. I think OpenClaw feels like you're talking to like a friend in like Telegram and cloud code dispatch and the channel thing is like not super reliable. And, and if it's like a coding kind of first interface. So it doesn't feel like text messaging a friend.
A
Yeah, exactly. Yeah. Like dispatch is. You're like locked into their app and sometimes it can be good, but it also doesn't feel. You don't feel so free with it. I feel like. Yeah, same experience with the Telegram and the Discord plugin was, you know, causing a lot of problems for me. So I still prefer openclaw in that sense. Yeah, so that's the second one I would say is this feature of heartbeat which makes OpenClaw kind of feel more like this always on thing. So heartbeat for people that don't know, it's basically this like thing that every 30 minutes it kind of by default triggers and openclaw comes live like every 30 minutes and that goes through some tasks that you've specified inside of a Heartbeat MD file. So this gives it this always on nature. Now Claude code, they've built some things over the last few months to I think mimic that a little bit. Like they built this loops feature but it's kind of for a different use case and it also, it only lasts for three days and then it kind of turns off. And yeah, I think it's not really built for this Heartbeat functionality. You can kind of hack yourself into it a little bit. But the actual Heartbeat feature I think doesn't really exist yet in cloud code.
B
What kind of stuff do you put in your heartbeat? Just at a high level.
A
So I was trying different things and one of the main things that I had in there was actually just to like improve the memory. You remember like in the beginning the open claw memory was pretty bad and then I kind of just put into my heartbeat for to constantly like check the sessions and like what I've been doing on every heartbeat and just save it into my daily memory. So that was like one of the things then I put there also like management around my TO dos. So I wanted my to do list to kind of feel like this thing that's just like self checking itself off and working with me throughout the day while I work in my openclaw. So I had like something in there to say like hey, go check, you know what I've been working on and check it off my to do list by yourself. So I don't need to do it myself, you know.
B
Yeah, the memory still needs improvement. So yeah, it's good. I didn't put the memory stuff in the heartbeat.
A
Yeah and then there's whole thing around cron drops so I think it's still much easier to use that in openclaw. So that's definitely a pro. Like, you know, openclaw is like this supposed to be this thing that's either living on a separate device or like somewhere on a VPS in the cloud. And so I think it's much easier because of that to be able to like create cron jobs. You just say like create a cron job and then it just runs you, you know. And with Claude code they've introduced this. First it was called, I think schedule feature and now they renamed it to Routines. And it's a little bit complicated to use because it's like there's a local version that runs locally and then you can also do like a cloud hosted version. It's like a little bit more complicated to use.
B
Although it is nice to see like a list of your routines in the clocko app. Because OpenCloud, I don't even know what kind of cron jobs I have to ask it what kind of cron jobs I have.
A
Yeah, that's true. Yeah. Or you have to go into this, into like the really bad OpenCloud dashboard UI and try to find them there or like into this cron job file. But then it just looks like code. Cool. And then I would say the fourth point that I put in the pro of openclaw is this like, feature of sub agents that you can just easily create a sub agent and, you know, then can have your main agent interact with it. And I think cloud code doesn't really have that. They have this like. So there are a couple of ways around it. You can kind of use an app like Paperclip and combine it with cloud code and then you can manage sub agents. And they recently released this Claude managed Agents, you know, product, which I think is going in that direction. But right now it's not really consumer yet. I think it's more for developers.
B
This episode is brought to you by linear. When engineers use tools like Cursor, clock code and codecs, a lot of work happens invisibly. Someone can go from a bug report in Slack to a shipped fix without creating any record of what happened outside of the code editor. And that's fine for speed, but it makes coordination hard, harder. As you scale, Linear integrates with the very best agent coding tools directly, like Cursor and Codex. That way anyone can see what an agent is working on and who assigned them to the task. You get the speed of agents without losing visibility across the team. Product teams at OpenAI, ramp and block are all using Linear to collaborate with AI agents. And I use Linear myself to run my creator business. So check it out at Linear. Linear App Agents. That's Linear App Agents. Now back to our episode. What do you use sub agents for? Like, you have an Orchestrator agent and a sub agent for like updating your newsletter posts or editing videos. What do you use it for?
A
Well, to be honest, I actually, I know there are a lot of people on Twitter being like, I have sub agent for that and, you know, like for my content, everything. And I think for most people, they actually don't need sub agents when they use openclaw. They just need like their main openclaw agent and then they can create groups in Telegram or wherever. They use it to talk about different stuff. And that's totally enough. And that's also how I use it. I think sub agents maybe start to make sense if you really want to separate context. So if you want one agent for each of your businesses, for example, then it starts to make sense. Because otherwise you constantly have to manage context between all of these sub agents. Right. That also gets very, very complex.
B
I do think one thing I do use it for is like, if I draft a piece of content and I want someone to evaluate it or kind of review it, I have a sub agent or another agent for that so that it doesn't get biased by the fact that I drafted the content. So this is the reviewer and the drafter is separate.
A
Cool. Then let's go into the pros of cloud code. So I would say the biggest one right now for me is reliability, because I don't know if, like, you've experienced that too, but with openclaw, it's still. It just breaks randomly. All of a sudden an update comes out and then something breaks or like the model just stops working, you know, Or. Yeah, it just like it's just kind of still a quite techy product. Not super reliable yet. Yeah, so that's one thing. The other one is that, I mean, obviously when like Anthropic cut off the access to OpenClaw, I think that's when a lot of people switched over and a lot of people said that the experience of openclaw was only this good because it was using the Opus model. So I would say right now it's probably still the better model. And that's like one of the reasons to use Claude code. Although, I don't know. Have you tried the new GPT 5.5?
B
I have not tried. GPT 5.4 is terrible with OpenCloud. So like. Yeah, because of. Because GPT5.4, I started switching more over to cloud code, they said 5.5 is better. Just like a guy working on it now. But apparently you can also still use the cloud CLI in openclaw in some way, shape or form. But yeah, the previous model, it just did not work with OpenClaw in my opinion.
A
Yeah, it was not as proactive as Opus was. Yeah. Um, and then it's I think overall also more secure cloud code. Although, you know, it's like the security kind of shows itself in off very often in just like asking you to approve something which can also get annoying and. Yeah, and then like one way I've went around that is to kind of just set like bypass permission mode to always active and then I guess it becomes also a bit less secure.
B
But it's kind of. Because I think you and I have built systems on both and it's kind of like, it's almost like fake work to a certain extent, trying to figure out which of these systems to use. But right now I think I have all my email management calendar stuff and like, just like trying to update Google Docs and stuff still in openclaw. And then cloud code, I use it to just like build. If I want to build anything, I use cloud code and maybe Codex to build. I'm not sure how you're dividing things, but.
A
Yeah, exactly. I think overall, like OpenClaw is, it feels more like this. It has this potential to be more autonomous. It feels like this thing that you're supposed to be having somewhere in the cloud or somewhere on a different machine, and that's supposed to be more like an employee that's doing stuff for you autonomously. And to me, cloud code still feels more like a actual tool that I'm using on my main machine that's augmenting my work.
B
The reliability is a big factor, man. Well, I guess cloud code isn't that reliable because they keep having downtime, but assuming that they can fix that, then the reliability, like I switch a lot of my cron jobs over to cloud code just because it's more reliable. But now I guess now why don't we. You have a full demo for cloud code ready, right. So why don't you show us a tour of how your cloud code OS works?
A
Yeah, exactly. Yeah. Basically because of the openclaw being unreliable, I've kind of tried to build my openclaw system also in Claude, just so that I like basically have both running at the same time. So yeah, I built my this Claude OS system basically. And maybe before I jump into the architecture, let Me just show you, like, a couple of examples, so I think, you know, people can get a better feel for how it actually works. Okay, I'm first going to copy this one and just open up my cloud code here. And by the way, this is the new desktop app, which is really awesome. I'll just paste this in here. I'm doing a demo, so it shouldn't share any sensitive information. And then I'll take this first prompt, and the first prompt basically says, get the transcript from this video. So I'll go get a. There's like some video link, some Instagram video link, paste it here. So get the transcript from this video and write two similar scripts for my offer. Then save them as docs in this G drive folder called Claude Content System. Okay.
B
You can. You can get transcripts from Instagram videos automatically or it has some sort of tool to make.
A
So in this case, it has a tool here which is called talkscript. So it's an MCP tool here, and that basically helps it get the transcript. And then after it got the transcript, it. Yeah, basically use the cloud code intelligence to write the scripts and then it uses another tool to actually access my G drive and, you know, create the files and put it in there.
B
The other tool is just like. Is it just like a gwc?
A
Exactly.
B
Workspace.
A
I would say that's probably, like, right now the most power powerful tool for me when using, like, this, this system, it just unlocks so many things because I can access my G drive and do anything with it, basically.
B
Yeah, I never bought into, like, just using, like, local MD files and Obsidian to manage everything. I prefer everything still in G drive because, like, it's in the cloud, I can use it on my phone and so on.
A
Yeah, I agree. I was kind of trying out both. And when I was building my, like, content system, I was, you know, having most of the things locally, but at some point I started realizing that it's just not very nice. Like, if everything is living somewhere locally on a machine and it's just harder to access. So, yeah, it's much nicer when it's in a. In a G drive. Okay. Okay.
B
It looks like it's. So you use this flow to basically make new Instagram reel scripts.
A
Yeah, I can obviously kind of define the prompt a little bit better or, like, put it into a skill for it to reference other files. But I just wanted to show how fast you can basically just get something from some website and then upload it into G drive. So where did you put this? Probably in scripts yeah, right here.
B
Awesome. And I think another benefit of this, like, for my open claw, like I don't give it access to my entire G drive. It has a separate email that I share some files with it because I'm worried it'll screw things up. I think with cloud code, it's just using my main credentials.
A
Yeah, you have like more trust in cloud code, right?
B
Yeah, I have more trust. Maybe I'll screw up too, but like, yeah, so far I have more trust in cloud code.
A
Yeah. Let's show this one. So this one says find, find all of the APIs I have used before and organize them in a Google sheet and upload to my drive. So yeah, this will do like a search into my local file system of all of the memories I've saved over time inside of cloud os and then just organize it, create a Google sheet, and then upload it into my drive again using gws.
B
Do you have some fancy thing for the memory? Like do you have Toby's QMD thing or is it just like doing searches right now?
A
It's still pretty simple. I kind of just mimic the system that OpenClaw has or probably even simpler than that, than what OpenClaw has. And it's been working fine for me so far. But yeah, I do think that at some point I'll probably hit some limits of it working well if it gets to a lot of files and very big files and then I'll probably use the qmd.
B
Yeah. Okay, so you actually set up like the daily summaries and the memory thing that OpenCloud has here. You just told it to do it.
A
Yeah, exactly, exactly. So I have like a memory MD file and then I have like memory, a memory folder with daily memory files in it.
B
Okay, you already are recreating an open claw. Yeah, exactly. I think Claude might have his own memory thing too. Like at the end of the everyday. It might save it somewhere. I just don't think. Maybe it's not in Claud code, maybe it's just in regular Claude. But I heard him talk about it somewhere.
A
Yeah, yeah, good question. Yeah, I think it's in the regular Claude sheet is ready. So basically it extracted all of the APIs that I was working with in the past days and just put it in here.
B
So These are the APIs that your Claudia OS has access to?
A
Yeah, exactly. These are all of the ones that I've been kind of either mentioning or like checking out in the last few days. So it probably took these from my memory files.
B
Hopefully you're not paying for all of them, but. Yeah, that sounds good. Yeah. All right, dude, so why don't you show us how all this thing is set up?
A
As you mentioned, my Claudio is called Claudia, and basically there's the UI layer here. So I like, mostly just use it through the new cloud code desktop app. And it's just like they've made such an improvement. I think it's been out for like 10 days. Before that, I was mostly using it through the terminal, but now that the desktop app is out, I'm just using it there. And I don't know if you've seen they have like these, you know, you can open up all of these things on the side as well. Like, you can see the files, you can, like, see a plan and so on. So it's really good.
B
Yeah. And they let you edit a markdown here, too. I don't know if you realized, but yeah, yeah, it's pretty good.
A
Yeah. So that's the main way I use it now. Terminal, I think, only sometimes if, like, a couple of features are still missing or like, sometimes I need to check something and then I'll use the terminal. And then if I really want to, like, go very into kind of editing a document mode, then I'll actually use something like Cursor or VS code with the cloud code extension. Because there I often want to, you know, I'll want to, like, type something into the document and then like reference something. So, you know, basically highlight it and then edit it with AI. And for that, I think having this extension inside of Cursor VS code is really good.
B
So you're using Cursor to just edit documents. Yeah, that's very funny. That's very funny. Because it's meant for coding. Right, but you just use it edit documents.
A
Exactly. Yeah, yeah. I think I haven't used it for coding in a while. I just kind of use it like Obsidian, basically. Okay, so that's the UI layer. And now the, I guess, important part is how is the folder structure actually set up? And I mentioned this before when I was setting this up, I was basically looking at how OpenClaw has done it and like copying that over. And so there are some, like, these instruction files. Basically. This is the most important one. There's the Cloud MD. It kind of replaces on OpenClaw side the Agents MD. And it's like the first file that is like being read and that has. Is like kind of the master instruction. It's like the system prompt. Right. So that's the important one inside of this. I can Actually show it. You can see here is like your Claudia, this is your workspace. And then I start with these files are imported into every session and then I reference these other files which I'm going to go into now. But yeah, basically I use this Claude MD to like pull in the other files as well. And then I have this part which is like a memory loop. This is to make sure that the memory here works. So this part basically says like it should save the daily memories. So after every interaction with it, it should save like one line into daily memory and then also sometimes save long term memory.
B
After every chat message it should save something.
A
Yeah, so I think one of the downsides of it might be that it, you know, makes every chat a little like, you know, slightly longer. But yeah, I didn't want it to lose any information, you know, so.
B
Okay. And then the identity and soul is just kind of the personality, right?
A
Exactly. Identity soul and then user MD is information about me and then various folders for business context. And then I think this one Tools MD is super important. I have also like every time I add a new tool then or you know, or a new MCP or CLI or something, I just say add this to your tools md and this way it knows like what tools it has access to. And I think that's the most important part actually. Then there's the like memory system. As I mentioned, I also have this like little thing built in now that it like overnight it does this dreaming. So it goes and checks the daily memory files and then like, you know, creates a compressed version of it inside of the long term memory file.
B
Is that like a cron job?
A
Yeah, it's a routine.
B
Yeah, but how do you prevent this thing from becoming super long? Like the memory md or is it meant to be super long?
A
Yeah, I mean my prompt is basically like, you know, just make it one line or something. One or two lines. So far it's not gotten super long. But yeah, I think probably at some point I'll run into some issues and then I have to think of a better memory system.
B
Got it. Okay.
A
Yeah. And then obviously the ENV to store all of the secrets. Like whenever I get API keys or things like that, I'll put it in here.
B
How do you share this is a dumb question. How do you share secrets with Claude? Do you just paste it directly into ENV or do you. Because you don't want to put the secret in the chat, right?
A
Yeah, I usually just open up my ENV file and put it in there directly instead of going through Claude. I think one of the other ways that I also have on my list of wanting to do is to basically have, like, everything in 1Password. I've already set up my 1Password to have a separate vault just for agents, and then I can share my passwords into that vault. I just need to make sure that my claudeos also has access to that vault.
B
Okay. And just to summarize, so basically every single conversation you have with Claudia, it runs Claude MD and Claud MD refers to these other files that it can reference.
A
Yeah. This way it then keeps building its memory and remembers things over time.
B
And for those of you watching, I guess we'll link to this. We'll link to these diagrams in the comments if you want to check it out yourself. Yeah.
A
Okay. I think this part is the most important part, actually. The tools, like, you know, there are different ways for it to use tools. There are different types of tools. Like, one of. One of the categories are CLIs. These are basically, you know, like little programs running on your computer. And I think, actually the fact that cloud code can use CLIS is one of the advantages over Cowork, I believe, because Cowork is, like, kind of sandboxed every time you use it, then it can't really use clis, but Claude code can. And that's one of the reasons why I prefer using cloud code over Cowork.
B
Yeah, I don't use cork. Yeah, there's too many buttons. I prefer just having a chat.
A
I mean, I understood the reason for people to use Cowork when they had only the Cowork desktop app. And Claude code was like, you had to use it through the terminal. But now that there's a nice desktop UI for cloud code, like, what's the point in using Cowork?
B
Yeah, I don't see a point. Yeah.
A
Okay. And then, yeah, there's obviously mcps, and I think, like, you know, there are so many different ways to actually connect your mcps. It's like, I don't. It's gotten pretty confusing at this point. Like, you can connect them through the desktop app. You can like, actually, you know, connect them directly from your cloud code, like in the cli, and it's just. I feel like it's a bit of a mess. But, yeah, at the end of the day, like, you can connect mcps and then your cloud code can use them.
B
Do you have them connected and turn on all the time? Because it's going to blow up your context. Right. Or do you have.
A
I do, actually, but I think it's gotten better. They've made this improvement that it doesn't, you know, bloat up your context that much. So I haven't really noticed a big difference.
B
Okay, maybe I'll give another try.
A
And then Obviously there are APIs. So, you know, nowadays when I look for new software tools, I just always, like, the first thing I look for is, do they have a cli? And then do they have mcp? And if not, then, you know, just at least like, do they have an API? And if not, then I try to look for another tool that has one of these.
B
Yeah, you kind of have these three things. Otherwise people are not going to use your app.
A
And then of course, there's this other layer of skills. Like I have, you know, I've created several skills for my own workflows. Basically, whenever I do a workflow several times I'll just say, like, turn this into a skill and then it's saved for future use.
B
Let's go through some. Let's just talk about some of these skills. Maybe you can describe a couple of these skills and what they actually do.
A
Okay, so this one, this Raven Grocery, is actually a skill that I've created to use the browser. And I created this first on my, on my openclaw as a skill. And I've basically like trained it to use the OpenClaw browser and then to like, log into my grocery, like, shopping application. And like, the first step is it takes what I've ordered last week and just puts it into the cart again, because these are like the items that I order every week. And then it just takes a list of any new things that I need that I added through the chat and goes and searches for the products and then adds it to the cart and then lets me approve it at the end. So it's basically like helping me automate my grocery shopping.
B
Okay, that's awesome.
A
This one is super useful. Video upload workflow. So I have this in my, in my content system where it's like I used to, whenever I upload a video for my editor, I would like go into Google and create these folders and then drag them in there and then get the link and share it with my video editor. And this workflow basically just does everything. I just need to say, run the video upload workflow. It creates all of these folders, and then all I need to do is just upload my files into exactly that folder and that's it.
B
I see, I see. Okay, I see. And what's the difference between project and user so project is in Claudia only and user is across all your projects, right? Your cloud code projects.
A
Yeah, exactly. Yeah. This one is only in Claudia, this one is global.
B
I think we're talking about this next. But like I feel like the whole podcast video production flow, like I just wish there was more like APIs and skills to automate this stuff. Like everything from recording the video to cleaning it up, to uploading it to YouTube to, to making clips, to making captions, making the show notes. All this stuff is super repetitive and I wish it was a way to just make it easier.
A
Yeah, I think the problem to that is that everyone that creates content has their very own custom workflows. And so it's very hard for a SaaS tool to come in and be the tool to use for everyone. But now that you have these tools with like, you know, cloud code and, or openclaw and connect, like being able to connect to all of these different tools, you can basically build these automations yourself.
B
Yeah, with the right APIs. Yeah, maybe we can talk about some of that later. But let's talk about routines real quick first.
A
Yeah, so routines are like new feature of cloud code where you can automate stuff. And there are. So there are two different types of routines now. One is local, one is remote. So local is you. Yeah, you always need to have the desktop app running, so your computer needs to be on. And then once you have a, like one of these local routines triggering, then it will actually run locally on your computer. So it can, you know, it's able to use the clis on your computer and it's just basically able to do everything that you're normally able to do when you're chatting with a tier on your computer. And then remote is kind of, if you want an automation to just always be running even if your computer is not on. And this one is a little bit like slightly more complex to set up because what you need to do before that is to, you need to create a repository and put all of the stuff into that repository that it needs to be running this automation. And then you like basically host, host it on GitHub and that way it can like run in the cloud.
B
So it looks like you only have one remote routine YouTube monitor. Right. So it's looking at your stats or something.
A
Yeah, this one is actually, it's like, it's looking at channels that I specified and then it's like scraping all of the latest videos of those channels and putting it into a sheets. So I can show you here. It's putting it like into this sheet with the, you know, like the views and likes and comments that it got. And this is running once a week.
B
And this is just to, for, to see what kind of videos similar channels are making. For inspiration, I guess.
A
Exactly.
B
It's, it's the, the titles are just like looking at the title, like just being a YouTuber is so easy to just go crazy, crazy like it's insane. It's, it's like I try not to do that stuff but like it's just hard because like when you write it's insane. Like the click through is like 2x. So you probably have another skill to do. You have another skill to analyze this stuff and maybe give you some suggestions or something.
A
Yeah, so I have this one skill that I think it's still running on my open claw, but it basically takes all of the things that I have in my idea folder, like all of the manual ideas that I've logged as well. We can go into that also later in the content machine. But yeah, it takes all of the ideas and then turns it into a weekly plan.
B
That makes sense. I think the problem with AI stuff is there's new stuff dropping every week. So some of the stuff that you've pulled is already out of date.
A
It's like, yeah, okay, sure.
B
Let's talk about the content system. Yeah, this is the capstone of this whole thing.
A
Yeah. So basically once you've set up, hooked up the tools and set up this foundation, you can start building systems like these, where this system is essentially like a collection of a bunch of skills, a bunch of like little automations that are like, you know, coming together to build the system. So this one specifically is to create short form video content. And I built this initially in my Open claw and I'm kind of in the process of migrating some of it over into my cloud os. But the way I, I've been operating it is mainly through Telegram. So I have this like Telegram group here with all of these subtopics and then I for example have this one for ideas and this is for the first step. So there's this idea capture step where I have several different ways to capture ideas. So one of them is directly through my telegram. So I can just say, you know, if I get an idea about something I can just say compare openclaw with cloud code and it basically logs that as an idea in one of my idea docs. And then I have the, the YouTube scrape that I showed you this like automated way. And then I've set up another way where I can. When I scroll through my Twitter and something, a post comes up as a good, you know, idea for inspiration, I can just like send that to my OpenClaw bots Twitter account. And then I have an automation running which like looks into the DM and just logs all of these things that I've sent it.
B
That's very smart. Okay, so you just DM your open cloud account. Got it. That's very smart.
A
Yeah.
B
Because I think I've been bookmarking it and I don't know, I don't think there's like a bookmark API.
A
Yeah, exactly. Yeah. I was thinking like, you know, I bookmark so much stuff and it just lands in the bookmarks graveyard, but there needs to be a way to get it out. Okay. So once the ideas are captured, I go into like a weekly planning step and there essentially it just like takes all of the ideas and just creates a weekly plan for me. So it picks out some of the topics and then writes like, puts it into a schedule, Monday, Tuesday, Wednesday and so on. And adds a little bit more detail. Yeah, this one relatively easy. But then the next step is once I've approved this plan, it can then start writing the scripts for me based on my past scripts also. So I've built up like a library of like my past scripts. And if I go here into. Yeah, here. So here's like. Yeah, yeah, so here's you know, like a content folder and it has short form scripts and it's like all of my scripts are living in here. So it creates like one script for each day.
B
Wow. Okay.
A
And then I can obviously go in and like review them and add like a bit more detail if I want. And then it goes into the filming step where I just take that script and put it on my phone and basically like talk the script into my phone. Yeah, like some people I think can, you know, they can just like talk into the camera and they don't really need a script or maybe just need like some bullet points. But I usually need like at least something to talk about. I need to see something. You know, I can't just like open my phone and just start talking.
B
That makes sense.
A
Yeah. And then after that I have a video editor in the loop of this process. Like you could also do it without a video editor here. Depends on the style of your videos, of course, but my style of videos, it's usually like me talking and then sometimes it's me filming something on the screen to like show something. Yeah, and like you could do that without the editor. By just like talking and then turning it around and filming and so on. But yeah, the way I do it is I just film myself and then I film the thing on the screen and then I just like get all of those files and just send it to my editor. And this process here is also like semi automated where I just say like run the video upload workflow and then it creates the right folder. It like pulls from the actual script so that it knows what it's about to create the folder name, creates the folder and then I just need to basically from my phone just click the folder and upload everything there.
B
You record and upload from your phone?
A
Yeah.
B
Why don't you use some, you know, video recording thing on the desktop? It's just easier I guess. You're making clips so it's just easier to make from the phone.
A
Yeah, true. And I also, I like this natural style of, you know, this like UGC content on the phone. So you know, I've, I've tried like different kinds of content and usually the like most natural style just performs the best. So I kind of just like over time just went with that.
B
Yeah, that makes sense. Just like off the cuff kind of thing.
A
Yeah. Once that's done, my video editor will send me the link back to the drive where the finished video is and then I can just paste that link to my agent and my agent can then use a tool, a CLI tool to like post it on the various platforms.
B
Let's go. Because I'm still manually posting to type Foley and then scheduling. So what's the CLI tool that you use? Post.
A
So this one is post its. Yeah, I think they like, they really exploded and mainly because of openclaw and their cli. So yeah, their CLI is pretty good.
B
So this is for posting video specifically. Right. It's not posting to X or anything. It's like posting clips.
A
I think actually they might also have X integration. But yeah, I use it for video for posting to YouTube, Instagram and TikTok. Yeah, you can add your captions and so on there too. So I actually have that also like partly automated where it generates the caption based on my script. But yeah, what I was going to say is that you want to be, especially with TikTok, you want to be a bit careful with posting through API. Like I think with Instagram and YouTube it's usually fine, it won't affect the performance. But with TikTok it's, it might affect the performance a bit. And also if you're starting a New account. You kind of want to warm them
B
up a bit first with like, human uploads.
A
Yeah, with human. Yeah, exactly, human uploads. And also with Instagram. Actually, that's one of the things why I was also still posting Instagram manually for a while and is like they, they have this edits app and at some point, like in the creator circles, people were saying if you upload it into the edits app and then from there you post it on Instagram, then it performs better. And I did that and I think it worked. Like, it's always hard because you can't a b test it, but I think it worked.
B
Interesting. Okay, I'll check it out. Yeah, and then you get your stats. That's the final step, right?
A
Yeah, yeah. Actually, one thing here that I also find pretty useful is, you know, in Instagram you often do these videos where you can give away something like a resource, you know, like comment for this resource. And I also have that part automated where it's like, it's based on the script. It creates a notion resource with all of my links inside of it, and then it just puts that link of the notion resource into my manychat automation. And then that part is also automated. And that also saves a lot of time. Wow.
B
All right, I'm gonna have to digest this a little bit more. I think all of us do. But it sounds like the most manual part is just like creating the video and actually editing the video. That is the manual part, right?
A
Yeah. For me personally, the manual part is like here, because I want to make sure that the script is good. And yeah, if I just let AI write the script like completely, sometimes it works, but like very often I just want to make sure it's good and go and do some manual edits there. And then this part is actually not. The filming part is not taking a lot of time for me because once the script is ready, it's just like 10 minutes, basically, like you open your phone just like say the script or like act the script, more or less.
B
Yeah, yeah.
A
And that's it. And then, yeah, the, the video editor, you know, but I. I think he probably is like, you know, one video maybe takes him like half an hour, short form video. So also not that much time.
B
Yeah, that's what I was going to ask you about because you wrote in your title is the no AI slob flow. And I was going to ask you if you just take the script directly and just like, you know, record it. So it's good to hear that you don't you don't just take directly from AI.
A
Yeah. The workflow that I like the most is I have it based on the planning. I have it create like some notes. So the script is actually not a script, it's more like notes. And then I'll have a look at the notes and just like, use my whisper flow to add, like, more points into it. And then I have a skill which basically transforms notes into the, like, script style that I usually want. So I'm basically, you know, adding some notes, transforming it, adding notes, transforming it for, like the seven videos that I post. Got it.
B
And through this flow, you're able to post a video a day or something? Like a clip a day?
A
Yeah, yeah. Roughly.
B
Yeah. Awesome. Dude, this is super helpful. Yeah, I feel like being a creator, like, we do so much repetitive work that, like, I feel like creators are kind of at the forefront of AI adoption because, you know, usually we don't have a big team and there's like so much work to do.
A
Yeah, I agree. And I think there's so much more you can do with this, right? You can, you know, this is like only short form content. You can hook it up into like long form part and like repurposing for Twitter and your newsletter and so on. So there's a lot you can do.
B
All right, Morris, well, thanks so much, man. This is super helpful. Do you have one final, like, do you have a closing tip for someone who wants to build something like this? Like build their personal OS on cloud code? I think the tip is maybe just like, build it one step at a time. Right? Like, think about what's taking up most of your time and build it one step at a time.
A
Yeah, exactly. Just get started. I think the most important part is getting like this initial structure in place, the folder structure, and then once you have that, just like starting to connect your tools one by one. And, you know, you don't have to connect everything at once, but, you know, just start with one and then, yeah, go from there.
B
Awesome. And where can people find you online or your stuff?
A
Yeah, on Twitter my handle is moritzkremp, so yeah, you can find me there.
B
All right, Morris, thanks so much for walking through this. This is super helpful. Pretty advanced, but I'm sure a lot of people will try this.
A
All right, thanks for having me on.
Episode: Build a Claude Code Personal OS Step by Step in 40 Minutes
Host: Peter Yang
Guest: Moritz Kremb
Date: May 10, 2026
In this episode, Peter Yang interviews Moritz Kremb, a product creator and automation enthusiast, about how he set up a fully functional Personal OS using Claude Code, inspired in part by OpenClaw’s agent paradigm. The discussion opens with a head-to-head comparison of OpenClaw and Claude Code for building personal "chief of staff" automations, and then delves into the specifics of Moritz’s Claude OS setup—including folder structures, memory handling, practical workflows, and automation routines. The conversation is highly practical, showing listeners exactly how to build up a system step by step to automate recurring creator and business tasks efficiently.
"One of the standout features [of OpenClaw] is you can bring it into a huge range of different chat applications... makes it so popular." —Moritz (01:21)
“…get something from some website and then upload it into G drive… super fast.” (14:50)
“I have… a memory folder with daily memory files in it.” —Moritz (16:49)
End-to-End Video Production Automation:
“When I get an idea… I just say ‘compare Openclaw with Claude code’ and it logs that as an idea.” —Moritz (33:21)
“I have that also partly automated where it generates the caption based on my script.” —Moritz (37:26)
Human-in-the-Loop:
“I want to make sure the script is good… very often I just want to make manual edits there.” —Moritz (39:36)
“Just get started. The most important part is… the initial structure. Once you have that, start connecting your tools one by one.” —Moritz (41:57)
| Timestamp | Topic | |------------------------|------------------------------------------| | 01:09 – 11:52 | OpenClaw vs. Claude Code - Feature Showdown | | 12:12 – 17:02 | Live demo: Claude Code OS in action | | 18:35 – 22:40 | Folder structure, memory, and config | | 22:40 – 29:05 | Integrating tools, skills, CLIs, APIs | | 29:05 – 30:47 | Automation routines: Local vs. Remote | | 31:50 – 41:09 | Content system automation, end-to-end | | 41:28 – 42:16 | Closing thoughts & practical advice |
To build your own Claude Code Personal OS:
The episode is highly recommended for anyone serious about workflow automation or building their own personal “chief of staff” with modern AI tools.