OpenClaw: Paradigm Shift or Slop Hype?

Hello, and welcome back to Freedom Tech Friday, a weekly live and interactive show hosted on the Ungovernable Misfits, X, Nosta, YouTube Rumble, and Twitch feeds. And I think we do have a full house this weekend, gents. So it is a momentous occasion.

We go live for one hour every Friday at 9AM eastern or 2PM UK time. And you can, of course, catch up later on the Ungovernable Misfits podcast feed. On Freedom Tech Friday, we like to cover the latest news trends for anything relating to Freedom Technologies.

That could be anything from Bitcoin or Monero, encrypted messengers, privacy tools, and everything in between. Essentially, if there's a news item, tool, or topic that you that can help you take back some control in today's digital Panopticon,

we wanna talk about it. My name is Q and A, and I'm head of customer experience at Foundation where we build Bitcoin focused sovereignty tools. As always, I am joined by my good friends, Max, the head honcho of the ungovernable Misfits Empire, and Seth, who is COO at CakeWallet.

Freedom Tech Friday is a live and interactive show, and you can help steer the conversation by commenting live. If the chat was actually working, it doesn't seem like it is, I'm gonna work on that, guys.

Pre submit your questions or topics on socials, boosting the show via podcasting two point o apps, sending in tips, or just sharing the show with your friends. Top support from last week's show where we covered off Apple and the privacy trade offs that come with the walled garden comes from none other than late stage Huddl, who sent 6,006

sats and said, quick boost before I forget or quick boost before I forget. You can read that however you like. Thank you for your support as always, late stage hovel. Without further ado, let's dive into the show. Mac, Seth, how are you going, gents?

Good morning, y'all. I'm doing pretty well. Doing pretty well. It's been a good week. Everyone's healthy. Good topic today. Gonna talk about some fun stuff, so I'm I'm doing well. Me too.

Hoping you tell me that this claw bot can do all the stuff that I hate doing, and we can all relax a little bit. So I'm looking forward to finding out. Yeah. I mean, if you don't wanna stick around for the next sixty minutes, basically, the TLDR is you can all retire. You're not gonna lose your jobs.

It looks like the chat's not working yet again, guys. Just to be clear, we can see your messages. Well, I can see your messages coming through in the back end on Restream. I've tried turning it off and on again. I can imagine Jordan sat there tearing his hair out right now as am I trying to fix this, but I have no idea

why they're not working.

It is yeah. Why doesn't show for me either. So Interesting. Yeah. I can't see it in the studio either. I'm you may see some random things pop up in a minute, guys, while I try and fix it. Oh, here we go. I see it now. Oh.

Okay. That is it. Yeah. That is it. You just don't see the images. It looks a bit different than normal. Oh. Oh, god. S x six. Here he is. Fuck. Okay. Think we're in business. It seems absolutely massive compared to normal.

Put me in There we go. Look at this. Jordan, all faith is restored to me.

Done, top of the pile. So, yeah, well, we both had busy weeks, guys. How how are we both?

Yeah. Nothing too crazy. We got a big release we've been working on, so we're trying to nail that down. Bunch of the teams out in Mexico City from Aerotopia, but I I stayed home. So I'm enjoying home.

Nice. Nice meeting you. Teasing on the what's in the release. Any anything big worth noting? Both of y'all know the majority of it, but, yeah, the the main teaser and Vic's already leaked all of the alpha, so there's not a whole lot He's doing that. I have noticed. He does. He does. He's our our resident and intern as well as CEO.

Biggest things are Lightning finally, and then new UI from the ground up that we're we're finalizing. So there's a lot of work going on behind the scenes, testing that, finding all the issues, improvements,

getting that ready. But lighting, honestly, has been ready for a good month or two now. Just been trying to finalize this new design that will pair well with it and give people a a whole nice new cake.

It's I've I've had a a a sneak peek behind the curtain, and I can tell everybody that's listening to this or watching this, it does look incredible. The new UI is is chef's kiss. So hats off to the well, to the whole team. Yeah. I'm looking forward to to this one being public.

Max, what about you? Busy week. Have you been doing more woodwork or been dealing with sick kids? I have actually. I've been doing some more stuff on the house. No sick kids. I've been messing around with AI lots.

Not not OpenClaw, but just, like, image generation stuff, making some more videos for different shows in the Uncoverable show list, doing all sorts of different artwork and bits and pieces, trying to clone voices, do all sorts of stuff and playing around with AI, but just still in the in the dumb old way, which is taking a lot of time.

Glad to hear it. And and I am I wouldn't normally do this such at the top of the show, but Lake Surfer has just stopped us on Nosta with 21,000 stats. Oh, you legend. See, calling Chicago misfits.

We are in the process of putting together a Bitcoin coworking slash restaurant slash event space. If you wanna get involved, check out Foise, spelled f o I s e, on Nosta for events. Well, thank you for your support. I know it's a soft ad, you know, that's fine by me. That's alright. 21,000

set ad is is reasonable on here. I think that's very good. Thank you. Yeah. Well received. Well received. Okay. Well, we've digressed for too long already as we always do. Okay. Today, big one. We're we're talking about arguably one of the most hyped, if not the most hyped open source projects in recent memory, OpenClaw.

If you've been anywhere near tech Twitter, GitHub, or any of the usual socials in the past couple of weeks, you've probably seen this thing absolutely everywhere. A 150,000 and climbing GitHub stars. I believe it's broke a record to be the fastest climbing or fastest rated GitHub repository in such a short short space of time.

2,000,000 visitors in a single week. Mac mini selling out because people wanna self host it. More on that bit later. The pitch is, you know, quite compelling. And on the surface of it, it seems seems to tick a lot of boxes.

Open source. Sorry. An open source self hosted AI agent that runs on your own hardware, connects to the messaging apps that you've already got, manages your files, and can run completely offline with local models with an asterisk.

No cloud dependency, again, another asterisk. No corporate middleman, another asterisk. And your data stays with you, final asterisk. Sounds like freedom tech. Right? Well, as with all of those asterisk, there's a lot to unpack here. The grass does seem that it's quite green inside this new ecosystem.

But in the last two weeks, security researchers have uncovered hundreds of malicious plugins, stealing crypto wallets, API keys, thousands of exposed instances open directly to the Internet. Basically, people just opening their computer up to people on the Internet, which is can enable all sorts of shit like critical

remote code execution vulnerabilities and even nation state groups circling the wreckage as well. So today, we're we're asking the question that matters for for anybody that's in and around this space.

You know, when an open source AI agent promises you sovereignty but ships with what seems like an absolute minefield of security holes, is it Freedom Tech? Is it a trap? Or is it somewhere in between?

Before we get into it, as always, we're gonna set the scene with what the hell is OpenClore in a minute. I'm just curious whether any of you guys have got to grips with it yet. I know you're both busy guys. Have you had a chance to to install it on your personal computer and give it all of your API keys?

I personally have not. Both time constraints, but more honestly, I don't really want an agentic AI running on my machine. Specifically because of I'll I mean, we'll get into a lot of this throughout the show, but the the two biggest caveats for me are that while OpenClaw itself runs on your machine,

the actual model that you're querying all this data against generally is not running on your machine. I don't have anything powerful enough here to give me the performance level I would want if I was to use this running the agent or the the model actually locally.

So then I would be sending all of this data off to who knows where, to ChatGPT, to Anthropic, to whoever. So in general, if I could run it against a local model that I'm actually hosting at home, I would maybe do it. But even then,

the combination of my cybersecurity background and privacy background are, like, throwing all the red flags at this type of thing warranting a lot of caution. So for me, I'm kind of letting it ride out a little bit, seeing where it ends up.

And I I definitely will want to try it, but, honestly, I won't touch it until I can actually run it against a local model so that I can have some privacy even if the security is still an issue.

Max, have you managed to break it yet? I would have if if I'd tried. But I'm sort of just holding back a little bit, watching what the smart people do, seeing how they get tripped up, and just saving myself some time this time. Because over the years, I've always jumped in

way too early with things and just wasted so much time and and pulled so much hair out that I think I'm just getting older and wiser and just learning to stay back. So desperate to do it, but similar kind of privacy concerns without knowing exactly what they are. It just feels as though

if I was to do this, I would not only fuck it up and break it, but potentially dox myself and cause myself major issues. So I'm I'm waiting. Okay. Well, I I have dipped my toes in very gently on a sanitized Linux machine that doesn't have Sparrow or any of my API keys or anything important just to to to kind of

test the waters a little bit. And it's been a frustrating experience as you as you might expect with a brand new open source tool. The my main problem we'll probably dive into this a little bit more after we've set the scene, but my main problem was that I was trying to do it on the cheap.

And I just quickly ran into rate limits for all of the free models that that are out there. So trying to do any any sizable piece of work, which is where, for me, the rubber really meets the road with this sort of stuff is offloading larger tasks that are, you know, kind of like admin based.

You still are gonna either have to have some beefy hardware like Seth alluded to earlier so that you can run a a decent local model,

or you're gonna have to pay for, you know, the an API service from from one of the the big few which comes with it, you know, all of the privacy trade offs and and obviously cost trade offs as well. But I'm gonna get that a bit later on.

So just to set the scene, what is OpenCore? I'll I'll kinda lead us off here. Created by a chap called Peter Steinberger, who I believe is very well off Austrian dev. He had a a very successful exit Karim. I think he did some sort of PDF reader document company that he managed to sell for, like, a €100,000,000.

So this turn this is basically a weekend project that he started back in November that seems to have blown up much bigger than I'm sure he anticipated. It was originally called well, it's had a a number of iterations.

It's been Claude Bop with a duck Claude as in claws of a of a lobster, I guess. However, Anthropic decided that they didn't like that and sent their lawyers after him and said, basically, you know, you you ain't calling it clawbot because it sounds like our trademark, essentially.

So then he had the terrible decision to rename it to Maltbot, which lasted for about twenty four hours and the whole of the internet told him he was a retard. And then he renamed it to Openclaw on I believe that was about two weeks ago now, and this is kind of like

the the name that seems to be sticking, thankfully, which is, you know, OpenClore is much better than Moldbot, and it kind of leans into the open source nature of of at least the the front end.

As I said, it's, you know, it's absolutely exploded on GitHub, and the the concept essentially is an autonomous AI front end. I I'm real open to call it an AI agent because as Seth said, like, you you need to give this an LLM,

a large language model to plug it into to do all of the lifting. The open source bit, like, people have been going wild for is kind of like a front end that enables you to speak to any of those models, and you can configure one or you can configure 50 different models depending on what you've got access to.

You can run it on your own hardware, literally anything from a Raspberry Pi all the way up to you know, some people are buying ten, twenty thousand pound Mac Studios to run this thing on.

It connects to your messaging apps, which is, for me, is one of the the key

developments here is that rather than having to go to, you know, chatgpt.com or claw.ai or anything like that and having, like, a a sanitized experience where you have to go to a specific place. Once you've got OpenClaw set up on whatever hardware that you have,

you can then configure it to message you on signal or telegram or anything like that so that, you know, you can be out doing the weekly shop with the kids and be like, ah, I wonder if I can just get my my AI overlord to get me to do some task for me. Oh, let me just message him on signal.

So it's and, obviously, they have the ability to respond there as well. So you can have that back and forth basically anywhere in the world so that you can do anything that you would normally do with AI, but from your phone. That's that's a a big part to me.

The the nature that this runs on, ideally, on some hardware that you control, it gate basically has god mode to whatever's on that computer. You know, that could be all of the files, shell commands.

It can access the web browser. It can do anything you can do, essentially. And if you then supercharge that with access to your Gmail and to your calendar, like, it is capable of doing one hell of a lot of things. Now you could view that as a good thing from a productivity perspective.

You could also see that as an absolute privacy nightmare, and I'm sure many people listening to this, myself included, share that viewpoint. As I said, it's model agnostic. Right? You can bring your own, you know, OpenAI

API keys, your cloud API keys, or you can connect it fully locally to to Olama if you've got hardware beefy enough to be able to to kind of handle this sort of stuff. So it very it really is kind of like a front end that enables you to kind of choose your own journey.

And then the final bit I'll pause on before I let you guys come in is that another part of the the kind of ecosystem here is they have the concept of plug ins. And there's already been, like, over 5,000 different,

like, skit they're called skills, but they're essentially just plug ins of additional kind of tutorials or pieces of software or or things that your bot can do tailored to a specific thing. So it could be, you know, the ability to crawl Reddit and do some research, the ability to go on YouTube and download the transcript,

the ability to create a full marketing plan based on some criteria you give it. Like, the this anything you can type into an an AI agent, you can turn that into a skill and have it as a reusable kind of tool for your OpenCLORE to be able to to leverage.

Any questions or have I missed anything glaringly obvious to set the scene for somebody who's been living under a rock for the past, you know, month or so, guys.

Not from my side. I think you covered that well, mate.

Yeah. Agreed. I think you I think you nailed the the core majority of it. Sorry. I was double muted there. It's it's honestly, like, I think interesting that it's so hyped, because it's not really anything new. Like, this agentic AI thing has been a

buzzword for, what, six months, nine months at least. Mhmm. But I think the the novel part of it is that now you can actually choose the agent, and you can choose what the agent can do, and you can connect it to the AI back end you want, which is a lot of the issues before. Was like, could have Microsoft Copilot,

which is in some way is a GenTick AI, but it's awful, and it's selling all your data from Microsoft and who knows who else. And the same with a lot of the other, like, GenTick AI that had been available is they're they're just proprietary black boxes that you don't get to choose anything about. You just choose whether or not you use it.

And a lot of times, they're deeply integrated into your system whether you like it or not. Whereas OpenClock gives you gives you that control. I I would hesitate to call it FreedomTech because the way people will normally use it wouldn't really be, but it it has a lot of the underpinnings of Tech and that it is open source and gives you

the choice and ultimately the responsibility for how you use it. Yeah. Absolutely. Alright. I wanna do a bit, you know, moving forward, a bit bit of a deeper dive on how you can run this because, like I say, this is just open source software at the end of the day.

It's a front end for you to be able to talk to any number of different AI agents and, you know, two people can use it in two very different ways and get two very, very different outputs. So once you decide that you want this and there's actually you know, it can do things for you that benefit you,

you can choose to have it live in a number of places. You can install it with a single command on any computer, laptop, desktop, single board computer that you have access to.

My strong, strong advice, and I don't think I'll get any pushback here, is do not install it on your primary device unless you are a very competent developer that wants to do some very complex things and you'd like to breathe fire whilst juggling knives at the same time.

Otherwise, do this on a dedicated machine. A lot of people have been leaning towards like an old laptop or a mini PC, hence the the Mac minis that all of the the YouTube influence influencer seems to be painting as the machine to run this on, which I believe is complete overkill.

But, you know, enough people seem to be following them because Mac Mini's are are selling out fast everywhere. Or a lot of people have also been running this on a virtual machine as well, which is, I guess, if you don't have access to a spare laptop

or desktop, then is is potentially a viable alternative. But you again, it comes with privacy trade offs that this is not your computer, and it can be turned off or taken over by whoever whoever's computer is actually, you know, managing that VM. So again, I would tread very carefully if you if you wanted to kind of go down that route.

Once you set it up, which again, to be clear, is a single command to install onto whatever machine you're you're accessing. You'll then be taken into a a t u I, which is a new terminology on me. I've not heard it before, but it it basically a terminal user interface.

Those of you that are familiar with, like, Sparrow server or Roan Dojo or anything like that, it's the same sort of thing where you have, like, a guided onboarding where you can, you know, configure it in any which way you want, tell it which kind of messenger you wanna communicate to it with.

You can give it all of your various different API keys for OpenAI or whatever AI you use. Or if it's on a a beefy machine with a local model preinstalled, you can point it at that as well. I'd be very keen to try that one out at some point in the future.

And then after that, you you're met with you know, you can either start communicating with this thing through the configured messenger, WhatsApp, Telegram, Signal, Discord, I believe, well.

You provided that you've given it, you know, the ability to speak to an AI agent somewhere, be that on your machine or in the in the cloud, you can then go and do all of the wonderful things that you would normally do with AI, which is kind of, like, open ended essentially.

Anything from, you know, help me fix the spelling on my blog post all the way up to help me write an app that rivals Cakewallet. You know, the the sky is literally your limit as to what you can try and achieve with this with this sort of thing.

Any questions on how you would kind of implement that, guys? Because I I again, is there anything you wanna double on to to kind of spell out for for the people? Main takeaway would be spare computer that you have control of the hardware on would be the strong strong recommendation here.

Yeah. My question is if you are gonna run it on your own hardware and you are gonna run a local model so that, you know, you you know where your data is, especially if it's for tasks that would matter if it got out.

What sort of computing power are we talking about to run something? I keep giving the same example, but for me, giving all of the images that we have, all of the artwork, all of the audio, all of the logins, all of the stuff that we do for social media, doing all the stuff that between us we do, and putting that on one computer

with it. So, you know, with everything local, what sort of money are you having to spend? Is a Mac mini actually what you need to run? My guess is it's probably that there's something else you could do much cheaper.

Like, how much needs to be invested compared to just saying, oh, fuck that. I'm gonna use my inbox and just pay them for a few months and see whether this thing works. Yeah. So two two clear distinctions here. You can run the OpenClore software on pretty much anything, even something as shit as a Raspberry Pi.

The the real question is where is all the heavy lifting being done? If you wanna do that locally for the sorts of stuff that you're asking for, like image gen, like, a Mac mini is not gonna cut it.

You might get away with something like a small version of of Olama on, like, a Mac mini where you can do some text based stuff where you'll probably have a little bit of a wait, but it will help you, I don't know, write blog posts or it will help you come up with some tweets to promote all the shows in your example, Max.

But if you start trying to do, like, clipping audio, doing image gen, like, you're talking you you need some big, big hardware here with probably with a dedicated GPU or something like that. Again, I I I'm just going off what I hear on on

YouTube videos on the topic. Like, you need some serious piece of hardware. Like, there there's people that are trying to achieve, like, Claude level or OpenAI level performance, and they have, like, two, three Mac studios chained together in, I think it's called, in parallel.

And they're, like, 10 k a pop. Oh, shit. Like Fucking hell. Okay. If you if if you're used to you know, you have the the 20 a month clawed friction where, you know, you can get by and you get snappy answers. It can do pretty much any of your tasks.

If you're used to that level of performance, trying to match that on something local is possible, but you're gonna have to pony up and the the the ROI is gonna be very, very

quite a long time. Now, obviously, price is only part of that because you've also got the privacy aspect of it as well. How much do you value that versus, you know, your $20 subscription and and letting OpenAI or Claude or whoever deal with or seal of the information.

Like, for you, if it's stuff related to to the show that's gonna be public anyway, I I would say probably just use the API keys. Use the the the most up to date model, and it's gonna be the fastest workflow.

If it's gonna be something like, I'm a self employed window cleaner and I need it to work out all of my taxes, here's my bank statements, like, I'm I ain't giving that to OpenAI. So the the there is there's lots of things to consider here. But if you're gonna run this locally, which I would love to do at what at some stage,

all things point towards, like, it's it's quite expensive at the moment. Seth, I don't know whether you've got any any other hot takes on that. Yeah. I mean, for the the normal stuff that a normal person actually needs to get done, as long as you're not in a rush, you could self host this on

something. I mean, I probably wouldn't go, like, less than a Mac Mini, honestly. When you can get a Mac Mini for, like, $400, I just can't imagine that you're getting something for less than that that you'll actually enjoy running Ollama on or dealing with any of this on.

But for that, I think, like like Q mentioned, anything that's essentially not image generation, you'd be okay. You'll just you'll have to wait a little bit. But a lot of like, honestly, a lot of the way that people use AI, that's totally fine. It could be, like you said, maybe it's, hey. This is the show title for this week.

Come up with a a tweet to put out on all different platforms and post it. That's not something that you need done in three seconds. That can be fine if you just send the prompt and you walk away or you start working on something else, and then you have it message you when it's done or something like that.

A lot of the general use cases, like, it's going to be drastically slower than what you're used to if you've been using ChatGPT or Claw directly or something.

But a lot times, that's just that's totally fine. You can go do something else while you have it do its work. That's the whole point of Agintiq AI is you get to do other things instead of the thing that you're having the AI do.

So I think it's generally, like yeah. I I think it would be fine for a lot of that. I probably wouldn't go below a Mac Mini unless you are really a glutton for punishment. Like, it gets it gets pretty bad when you don't have enough CPU or RAM to even keep some of it in memory.

But that's a pretty good yeah. Sorry, Seth. I'll just interrupt quickly while it's in my head. Could you do something where it's like a bit of a hybrid of both where with some clever permissioning or something, you could say, run everything locally other than image generation,

and I want you to do that using mid journey and then pull it back down. Like, you're gonna generate for me. Exactly what OpenClaw can do. Yep. Okay. Yeah. Because because that would be you know, I know you I can do image generation elsewhere, but I'm very happy paying the subscription on Midjourney.

Like, it does such a fucking good job. I'm used to it. And then my thinking was like, okay. This thing can probably get me, like, 80% of the way there, and then I'll go in and I'll do adjustments for most things.

But doing image generation of that quality locally, I totally get it. You're gonna have to be spending a fortune. But outside of that part, most of what we do is not that heavy. It's already done either by me or some other AI which we're paying for. So maybe maybe that's like a middle ground.

Yeah. I definitely think so. Go ahead, Kim. Go ahead. We yeah. We we already live in that world. Like, your your workflow is a prime example of where you can really leverage this sort of stuff. Like, most of what you create is going out public, so privacy is less of a trade off.

And Mhmm. You you can kind of slowly build up, it I guess, it would be classed as a skill where you could teach this AI to say, okay. We've just finished recording Freedom Tech Friday. Let's go through all the things that every week I want you to do. I want you to download the the the audio.

I want you to write a transcript from it. I want you to go through that transcript, and I want you to pull out all of the all of the hot takes and clip them up for me and put them into you know, so that we can turn them into tweets.

I want you to use my mid journey API subscription to go ahead and generate me thumbnail for the for the video and for the podcast feed and stuff like. You could do all of that after a little bit of training in a one bank from your phone on Telegram. Like, that world is is already here.

And if you know what you do, you can make that happen. Yes. You're gonna have to pay for it, but you're you're already paying for it. It's just that OpenClore is now gonna be like a single interface for you to be able to Yeah. Kick off that process.

Well, and you're paying for it in, like, the time that it takes. You know, we're talking about Freedom Tech. Freeing your time is a big one. Absolutely. Like Oh, yeah.

Just the shit the shit for one show that we do takes, like, twelve hours of, like, fucking around with, like, artwork and clipping and tweets and, like, all the and then editing, all the stuff, like, in the background. So even if it took 50% of that off per show, that's worth a lot to me.

Yeah. Absolutely. That and that's why I'm most bullish about this is is time saving for for menial task for me, like, both the stuff I do at foundation and prepping for for this show. Like, that's taken me significantly less time, and the output is is just as high, if not already if not better.

Like, researching topics, coming up with, you know, notes, notes, tweets, show notes, all that sort of stuff is just being becoming far far easier. And it gets better and better as as the AI learns more about my my preferences.

And I'm comfortable, like, sharing that sort of thing because, like I say, all of this is public anyway. We're literally on a livestream on YouTube right now talking about it. Like Mhmm. It's it's perfect for that sort of stuff. It absolutely yeah. I I I just love it for that.

And this is a while we're talking about this subject of, like, maybe doing it locally, t Jordan just posted that Sovereign Bitcoin, the Telegram group uses Snead, mentions that compute he he thinks computer prices are gonna aid actually to demand from the industry.

So if you're gonna experiment with these things, make sure that you make those purchases sooner rather than later.

Yeah. I mean, I don't know about eight x, but I can definitely see, you know, us heading in that that general direction. You can already see that by, you know, the Mac mini hype that's come out in literally in the last couple of weeks.

Just a quick question from the Nosda chat while we're talking about this sort of stuff, and I think we've probably already answered this, Max. But Observer asked, did you do local or remote AI testing for ImageGen and voice cloning?

Sort of. I tried no. I didn't try local, to be fair, actually. I looked into doing it and read a load of reviews and stuff, and, basically, it was terrible. And I did try other I did try other options outside of Midjourney, and they were absolutely disgusting. Like, Adobe's one was fucking horrible.

I find the one on x to be pretty shitty for for what I wanna do at least. What else did I try? I tried one where I pay with Monero. I can't remember what it was called now. Like, a paid per generation.

Pay per queue. Yeah. It was it was horrible. It was absolutely horrible. So for me, anyway, I was just like, no. This is it's just not worth my time. Like, if a lot of it is just, like, knowing how to use the tool.

And once you get comfortable with something, like people who use Adobe Photoshop, I can't imagine anyone wanting to put themselves through that torture because it is the worst interface I've ever seen in my life. But once you're used to it, people stay with it. So I guess it's a bit of that.

Yeah. Definitely. Okay. Alright. Let's get on to the the the elephant in the room here because this it sounds like it's like a productivity utopia, and it it can be for the in the right in the right context.

But as we alluded to at the the top of the show, there is a shit ton of risks and considerations that you should have if you're considering on running this software. There is a lot of kind of and this is not just a kind of theoretical.

There there have been so many leaks already of, like, different people's personal information. So I wanna get into, like, what should people be considering when they're looking to go down this rabbit hole. Seth, I know this was, a a a hot part of the conversation that you wanted to get. So I'll hand it to you first. But, like,

if somebody's never done this before, like, what what are the top things that you're gonna say, look, don't do this, do this, and be careful about this?

Yeah. I mean, a lot of this comes back to the same issues we've raised about AI in the past. Like, we've we've talked about this quite a bit in different different episodes previously, especially in the one with the folks at Maple AI.

To me, it really comes down to three main things that people need to think about. One, we've covered this already quite a lot today, which is privacy.

Where where are you sending all of this data to be processed? The default is sending it to Claude and trusting them with info about the most intimate parts of your life. Like, you're giving them access to your your email, to your calendar, to your messages,

they will know everything about your children. They'll know when you're going to the doctor. They will know that bill that you haven't paid and you're being harassed about. Like, they will know everything about you.

Even worse than a lot of us talking about the surveillance state kind of issue, think about being possible, because normally, a lot of this stuff is kinda segmented at least between different companies.

But now you're kind of you're and I'll I'll use a phrase for Meredith Whitaker from SIGNAL, because she's talked a lot about this very presciently months and months ago. But you're kind of breaking this blood brain barrier between your operating system and your applications, where normally there was kind of a divide that protected

each application from the others and made sure that no one thing could reach into all of the different parts of your life, to all of the different applications, to your signal, to your telegram, to your Discord, to your Gmail.

All of these things were disconnected, you had control over what to talk to what. If you choose to run something like this, everything that you give it access to, essentially, you should assume that the the company you're sending those prompts to,

that you're sending that data to, has access to all of that as well. So from a privacy angle, it opens up, like, a whole new paradigm where, like, Signal is a great example, and I I love that Meredith Whitaker is talking about this because it's a really good way to think about this. Signal is fantastic.

Intent encryption is unbeatable. Everyone knows that the the algorithm or the cryptographic approach that they use is fantastic, but the privacy is only as good as who has access to your Signal app.

And we've seen this where people add the wrong person to a group, and all of the privacy that they had is gone because they chose to add someone to that group. Essentially, of this as adding an agentic AI to all of your chats as a third party that's just looking over your shoulder, which again could be really useful,

but also can be a nightmare scenario where things that you think are private because they're intent encrypted between you and the recipient are also being sent off to this third party because you're asking it to summarize your messages or something like that. So that privacy angle is obviously one that you need to be concerned with.

The security one is another one that's very kinda closely intertwined where I mean, there's there's already been so many instances of malware being distributed in the actual the, like, the plays that you can run,

these skills that you get from ClawHub where you can download things that people have made to help you to do something more easily rather than writing it yourself or having your your local AI write it. You use someone else's skill that they've created so that you can simplify connecting into a specific app or doing a specific task,

which again is great. But if you're just downloading something from the Internet and then essentially running it with root permission on your computer, like, this is, again, worst case scenario for security. So you need to be very, very wary of what you're actually running,

that the things have a lot of installs, maybe even searching around to see if there have been any reports of vulnerabilities with this specific skill. Like, do your research before you allow these things. Because, mean, as you can see in the the quick start right there on screen, essentially, you're just giving this thing

permission to do whatever it wants in your terminal, which again, isn't necessarily a bad thing, but you have to be very cautious with with how you allow this access into your life. I think the the best suggestion so far for sure has been put this on a sanitized device that doesn't have anything else or if it hoses your system or

if malware gets installed that tries to do like crypto jacking and steal your your Bitcoin seed or something. There's there's nothing there for it to take. That's a great way to get started. But

honestly, if this is really gonna live up up to its potential, it's gonna need to be something that you can safely run on your main machine where you have all of the apps and tools that you need to interface with. So think those are the the biggest things for me, thinking about the privacy, thinking about the security.

And then, yeah, really just considering deeply what you want this to have access to before you actually just go crazy with it.

Some great call out

there from, like, I guess, a less technical perspective. Any any question that you'd have around, like, the the implementation on this and some other things that, you know, Seth or I might have overlooked because we we kind of take this stuff more for granted?

Yeah. I guess, like, my question would be or one of my questions would be, let's say you bite the bullet, you say, I'm gonna buy the hardware. I run this at home. I separate everything out so that it doesn't see any of my personal information.

Is there anything that people need to be wary of when connecting this machine at home? So let's say it's running on the same Wi Fi. Is that a problem? Should it be running a VPN on that machine, and can you I assume you can do that, but, you know, should you run a separate one to the VPN using on your other machines?

Do you need to do any clever networking stuff? When you start putting information on there that's specific to the task that you need, like images, audio, all this stuff. If you're sending that from your personal computer across,

do you have to consider how you send it? Is there metadata that's then gonna cause you problems within the stuff that you're sending over? Like, how do you do this in a way where you're not shitting on your own doorstep

and jumping through hoops that you think are useful and actually because you've forgotten one simple thing to do with networking, you've actually fucked yourself, and now your Bitcoin's gone.

Yeah. That's some some you raised some good points there, and I'll I'll probably defer to to Seth more on the the networking side of things. I I don't think that this could, like, you know, remotely hack into another of your devices because presumably they

or, hopefully, they're gonna be password protected. So, you know, it it still has the same protections against, you know, your phone, you know, remotely trying to get into your machine that's running Spiral or something like that. Like, I I I don't think that is a risk.

Could you do some more advanced stuff with networking to put it, you know, this machine on a segregated kind of guest network, then, yeah, absolutely. Is that gonna be beneficial from a security perspective? I guess, yeah. It's only gonna be additive.

It certainly can't do any harm. In terms of the metadata side of things, again, I unless you're using this 100% locally, I would just treat it as you because you're the the presumption would be that you're gonna be using a third party AI and just treat it with the same kind of

caution that you would as if you were just going to chat gpc.com. Like, don't put pictures of your kids on there. Don't put pictures of yourself on there. Don't give it your bank statements. All of that sort of stuff still applies even though this is, you know, I guess, in air quotes, running on your

local machine, which is true, but it's still calling out frequently for every single step over to the Internet. And then, I guess, on the the messenger piece, the only thing to to be aware of would be signal would obviously be the the best

option to be able to communicate with this. I probably wouldn't advise doing so via Discord or anything like that. And also be very careful around, like, how you kind of publicize the fact that you're running a bot. Like, if if somebody on Telegram finds your bot handle, then in theory, they can kind of send commands to the bot over

that from their Telegram account. And if there well, I don't know, actually. If it was in a group, then they won't be able to come out and sense. Yeah. I think maybe the API keys would prevent that. But if if you in unwittingly add this bot to a to a group,

then, you know, the you're gonna have a bad time if there's somebody who wants to have a bit of fun in that group. So be very, very careful how you communicate with this and what you do with the bot in terms of, like, how you know,

definitely don't add it to any group chats unless it's somebody that you or a group of people that you have a lot of trust in so that you because you're essentially giving them access to whatever machine that OpenClore is is running on. Can you imagine if it added itself to the Pledminer group or the Meshtadel,

The shit that would be on my Twitter the next day. Yeah. Yeah. Indeed. But, again, like, that's where it all comes back to running this on some hardware that you control. Like, just pull the plug. Like, if it goes tits up, just pull the fucking plug quite literally.

And that that's the power of running this on your own hardware where you're in full physical control of, you know, if it gets out of hand, then, yeah, you can handle it. Yeah. One one last quick call out on that. That I I know I said that there were three things, and then I only said two things.

So I brain farted and forgot the third. The third one is just, like, my my bigger and I've talked about this a lot, so I'm sorry if I'm beating beating a dead horse here. But my my bigger concern with this actually goes even beyond the privacy and the security angle where,

like, if people start actually leveraging this for large parts of their lives and just trusting the AI blindly to do the right thing, it's a little terrifying to me. I mean, anyone who has used a iPhone or has seen on Twitter the way that Apple's AI has summarized messages in some terrifyingly bad ways.

And that's like a and that's a very benign way that AI can screw up and make you think something terrible has happened when it actually hasn't and throw you into panic or just completely misunderstand the 100 group chat messages that were posted and get super confused and

do that. That's even just that's just you being misinformed by trusting the AI. But on the the actual side of, like, actually letting it post things for you or message a friend about something or handle some some business dev conversation that you have

getting back to people, like, the the miscommunications and the issues and errors that you can have if you don't actually validate what it's doing can be quite broad. And that's, like, just, again, another caution of, like, it can be incredibly useful, but also AI is really good at bullshitting. That's, like, what it does well.

And it could look on the surface like it's doing a good job when actually the messages that are being sent or the summaries it's giving you are just complete and utter bullshit, and it just wants to please you, so it just keeps giving you bullshit if you like that bullshit.

And I I know I've cursed a lot, but that's it's a it's a technical term when we're talking about AI. Yeah. Okay. It's a technical term. You're absolutely right. Thank

you for and then you correct it. Thank you for pointing that out to me. You're absolutely right. I was yeah. It's fucking annoying. But, yeah, you'd want it for review. I'd I'd want at least, like, please get all the tweets lined up, all the nostrils lined up, all the images lined up, all the, you know, clips lined up for my review

because otherwise, are gonna have problems. I'm sure you are. But that's still a massive lift even if you, you know, you're still making the final decision. It's just doing more legwork for you. And then you probably, let's say, have to change 20%,

30% of it, and some goes in the bin and some gets adjusted and edited, but it's still a massive time save.

Yeah. Absolutely. I think that there's always gonna be two ends of the spectrum, isn't it? We've already got people I've seen YouTube videos of, like, people putting their fucking credit card details in given that to come, like, book flights and stuff like that, which is absolutely horrifying.

I I saw one post on social again. I have no idea to be able to verify whether this is true or not, but, like, some somebody's open call went and signed up for, like, a $5,000 influencer business course,

like, just rinsed his his entire credit card balance Amazing. Overnight when he just let it Matt Brunell. I mean, for me, if that is true, then, you know, I I feel like that's a valuable lesson that that individual has learned.

We've also got the bots now. There's there's literally APIs being built for bots bot centric wallets where, again, they can have their own Bitcoin, Lightning, eCash wallets where they can go and do things with that on the Internet and pay all the bots. Like, this is a wild new,

I guess, paradigm shift that we're heading into. Yeah. And I think that there is always gonna be or should be that level of human backstop to be able to, you know, define, okay, these are the things that you can go and do without me.

Anything that involves spending money maybe over a certain threshold, you come to me. And, like, as long as that's all enforced within open source code, like, I would be, you know, open to the idea of doing that in the future because I think, you know, that that there is some value to be had there.

But just given these things like god mode to your main computer and given that your credit card details is a recipe for disaster and a and a place that I could never ever see myself getting into in any way, shape, or form.

Alright. So we do we do have some some questions. But before I go into the questions and there's a couple of comments here rather than questions as well, I'm gonna bring up that we can discuss, guys. Anything else you you wanna cover off in terms of, like, final takeaways?

Yeah. Just one more thing that jumped out at me. I was just double checking this because I was pretty sure that they'd announced it. Maple AI, and we've had we've had them on the show before and have have talked about them quite a bit.

One really cool thing that they actually released fairly recently, this was when was this? Actually, a few months ago, September now, is what's called Mabel Proxy. Basically, it's a it's a service that you can run either through their desktop app, or you can run it in Docker, Docker Compose, that lets you expose a OpenAI compatible,

like ChatGPT compatible AI endpoint or API endpoint that is actually pointed to your Maple account. So you'll have the good privacy that Maple has inherently. They're doing it in encryption with their prompts and getting the info back to you,

but you're able to essentially pretend to OpenCLOW that this is a ChatGPT instance. And so you can get a lot of the benefits of a remote AI that's gonna be a remote model or remote yeah. That's much faster than anything you could run at home, but still having really good security.

So that's honestly probably the best option right now if you want to try it and you don't have the hardware to run it at home, would be to to try Mabel AI. You would have to do their pro plan at least. Their paid plans are the ones, obviously, that that do have API access.

You can just do it for a month, or you can pay with Bitcoin and get a discount.

And that will give you a good user experience while still having good privacy in the day. And, honestly, that's probably how I'm gonna try it. I'll probably resubscribe to Maple and and try OpenClaw with that so that I don't have to actually run at home, but I'm still gonna have good privacy at the end of the day.

So just wanted to plug that real quick.

That's an excellent call, actually. Their their pro pro plan, what does that set you back? It's a $100, something, isn't it? It's 20 a month. So they have a max plan that is 100 a month, but their,

their pro is just 20 a month or 18 a month with Bitcoin. It's literally just for me. It's exactly the same. You get the same. You just pay more. Only if you're named Max. Yeah. Yeah. If you're named The Max. Pulled it up on screen. Just pulled it on screen.

Oh, nice. You go. It's a $100 a month, which is is that the the minimum That's the max. You don't have to get the max. To get the API access. No. Max was just joking. You can do the pro. Yeah. Pro and up all support the API endpoint that you need for running the Maple proxy. That's that's a

and once you're within that, can you choose the model? Ah, yeah. It says seven powerful models. Okay. So so you have a choice within that a that single API usage of whichever of those seven model models that you wanna use. Is that right? Yep. Yep. And you can choose that in the the Maple proxy

back end too. So you can choose which which model you wanna use in the day. They got those seven different ones. They're all laid out. I I dropped the blog post actually that they

they made about the Maple proxy, has details on like how to run it, what models are available, all of that in chat. So it's probably gonna be people's best bet if they actually care about privacy right now or self hosting something if they're a little bit more hardcore.

That's a really great call, actually. Because for me, the next step on this journey, other than just carving out more time to play around with it, is to go for some sort of paid plan with an API access so that I could start trying to stretch this thing's legs a little bit and seeing,

you know because I have no idea how far that $20 a month would go. Like, am I gonna burn through it in three days? Is it gonna last me the month? Is it gonna be more than enough? I don't have a good feeling for that at the moment. So for me, figuring that out is is gonna be the next step. And and obviously,

all of the different providers are gonna burn through tokens at different rates because they they all seem to well, that's the way they work. They they burn through tokens at a different rate depending on the model you choose. I'm sure there's a happy medium, listeners, watches.

If you have any, you know, private space with that and, you know, you're the best kind of balance between not spending hundreds of dollars a month, but also being able to get a good enough output from the AI that rivals, you know, what you can get from ChatGPT or Claude, please do let me know. I'm I'm I'm curious on that one.

Okay. Right. We let me just kill the screen share. We are almost at time. Let's go through a couple of quick questions. I do have a hard stop.

Where can I get a spare number to chat with it on signal, and how can I chat with it on Nosta? I'll answer the Nosta bit first. You can DM it on Nosta. Nosta DMs is one of the communication

preferences. You can basically, just when you go through the the TUI, the the terminal user interface, you'll have a list of all of the chat things. Not the d m is one of them. I haven't actually done it. Just be mindful that I think it uses the naive not the d m. So,

you know, yes, it's encrypted, but none of the metadata is, so it's probably not the best option to be communicating with. In terms of getting a spare number to chat with it on signal, probably something like maybe tech verified guys. You got any other better alternatives?

You you so you just have to be careful with Signal because you if you ever need to get that account back, you have to be able to access that phone number. So if you're okay, like, if you're literally using this only for

chatting with your OpenClaw and it's not for other messages and you don't care if that account just ceases to exist one day because you break your phone or something and you can't recover because you don't have access to that phone number anymore.

You could do that, but that's just the one caveat is to get back into your account. You have to have access to that phone number. So you could you could easily get yourself locked out of it if you use a kind of one time SMS verification service.

You could just have a whole another, like, phone line for that or find some something else. But, honestly, one that actually has SMS is a bit of a pain in the ass to find now nowadays, or you're gonna pay 15 or $20 a month for it, which is obviously painful for that.

So some other solution like Nostradiums would probably be better. Yeah. It's a little painful. But, I mean yeah.

Cool. Thanks. Alright. Let's wait through some questions. So two below on x. Oh, I'm a local agent or Maple AI. What's the best balance for privacy, but productive enough to make a difference?

If you can stretch to that $20 plan and you're not a heavy user, I'd imagine that's probably gonna fulfill most of your needs. I don't have any local experience, so I don't know, like,

whether just installing a small alarm or on, like, a generic laptop is gonna be enough for most people. I I don't have a good feel of that. But in my experience, I've been using Claude for work stuff just on the $20 a month plan.

I hit my limits quite regularly on that. I I would say as I'm more of a heavy user in recent weeks when I'm got getting it to do more research type of things. So as always, it it really depends on

what your balance sorry, what you're kind of asking of the AI, basically. If you're just using it as a qualified search engine, then the free one or or the the local stuff is probably gonna be okay. But yeah. What what about you guys? Any any other comments on that one?

Yeah. I think I'm I mean, I think we touched on it a lot already. I think Mayplay is maybe a Maple AI, man, is probably the best place to, like, try it. Like, do a month of Maple AI and see if you actually are benefiting from OpenCLL and if it actually makes sense.

And then if you really love it and you're hitting limits or something or you can do something cheaper, like, you don't wanna go to further max plans, $100 a month, and you'd rather buy a thousand dollar

souped up Mac Mini or something, you can always go from there. But, obviously, start small. I know there was a mention earlier of, like, buy whatever hardware you wanna play around with, but a lot of this stuff fades pretty quickly.

So just make sure there's actually use for you before you jump into actually getting separate hardware for this and everything. Yeah. That's a good call. It can be quite easy to get swept up in the hype of this sort of stuff and be be tempted to shell out a load of money. And, you know, this stuff is changing

literally on a daily basis at the moment. So unless you have a real need to be at the cutting edge, I wouldn't go shelling out thousands of dollars on new hardware just to play with this sort of stuff, you know, unless you you've got plenty of change in the bank.

Another another comment this was a useful comment actually from x for to one of your retweets of this, I think, Seth. He said this is summarized by the way, sir Jamsalot. So I have I've only got so much space on screen. But he said, basically, you've gotta choose between privacy or cost.

A seven k Max Studio lets you run models locally, avoiding privacy risk at a 500 plus per month API cost.

You you could finance it and own your AI long term. Go for a 128 gigs of RAM minimum, two fifty six gigabytes ideal, and the best models include Opus, Kimi, Quen, Coda, and GL m five. So, Jamsula, it sounds like you got some experience here. I might be sliding into your DMs, but I appreciate the insights.

And then one final comment from with a robot on Nos, they said, referring to me talking about how quickly you can burn through your tokens. He said if

OpenClaw is anything like Claude, which I guess it can be if you use the the Anthropic back end, You can burn through your time allotted credits on the 20 tier very easily. And that is my also my experience been using the the Claude desktop app for for for similar sorts of stuff. Obviously, it's all the same on the back end. So,

yeah, I guess to cap it off, guys, I would just say go out and play with some of the the free back end options. Obviously, if you're doing that, you need to be very conscious of the information that you're sharing with with the AI.

Other than that, the next step would be try and Maple AI on one of their cheaper tiers. See how far you can make that that monthly stretch. Get that plugged into your OpenClaw. Think that's gonna be my homework over the weekend, actually. I'm glad you brought that up, Seth.

And then if you start hitting your limits, then you're gonna have to start considering either, you know, ponying up for some decent hardware at home or probably the

the better option will be, you know, paying for a bigot here on one of the one of the API providers. Obviously, Maple being a great option from a privacy perspective. And, of course, you can play you can pay in Bitcoin.

Guys, any final comments before we we sign off? No. I don't think so. I think we we covered the bulk of this. Yeah. It's a it's an exciting thing. I definitely sound like a doomer whenever I'm talking about this stuff. Old man yell at cloud vibes,

but there's cool stuff about it. But It's important. You just have to proceed with caution. Yep. Yeah. Absolutely. Yeah. You did this countless, you know, countless feedback in the past couple of weeks of people literally getting wrecked with this sort of stuff. So you do need to be very careful.

Don't just chase the new thing all the time and throw away all your privacy. Max? Yeah. No. It's been good. I'm still not gonna dip my toe in yet. I'm sure we're gonna revisit this in the coming months. So I'm just gonna keep watching from afar and let the smart people do their thing. But thank you to everyone who's posted in

advice. I think was it? So John's a lot you just said about the the back and all that kind of stuff. That's all helpful stuff. So anyone else who has any advice, just just let us know. We'd appreciate it.

Yeah. Absolutely. Alright, guys. Until next week. Stay ungovernable.

Thank you for listening to Freedom Tech Friday. To everyone who boosted, asked questions, and participated in the show, we appreciate you all. Make sure to join us next week on Friday at 9AM EST and 2PM London.

Thanks to Seth, Max, and Q for keeping it ungovernable. And thank you to Cake Wallet Foundation and my Nim Box for keeping the ungovernable misfits going. Make sure to check out ungovernablemisfits.com to see mister Crown's incredible skills and artwork.

Listen to the other shows in the feed to hear Kareem's world class editing skills. Thanks to expatriotic for keeping us up to date with Boost's XMR chats and sending in topics. John, great name and great guy, never change and never stop keeping us up to date with mining news or

continuing to grow the mesh to Dell. Finally, a big thanks to the unsung hero, our Canadian overlord Jordan, for trying to keep the ungovernable in check and for the endless work he puts in behind the scenes.

We love you all. Stay ungovernable.