Free ai video creation + OCR full books + private photos + more Github finds

0:00Today, I've got two free ways for you to let AI make your videos for you. And I have two ultra-creative ways for you to save money on token costs. You're going to have a laugh, and you're going to see real savings. I have a free alternative to Apple Photos and Google Photos that will actually let you own your own photos and protect your privacy. You know what I'm talking about. And so much more today. I've got all the repos that are fantastic, but don't get enough attention. Links below, and of course, chapter marking so you can jump right into the section you want. [music] Let's

0:28get to it. Presented by Zapier, the AI automation company. The first one is Skill Specter from Nvidia. This is one that you wanted to have on the list, and you specifically asked for it to be the first one. Why? Reason is is because you cover a lot of different skills on this show, and I've gone through a lot of them, and it's like

0:47there's so much stuff going on in here. You can't tell whether there's anything weird or malicious going on within the side of skill that could be thousands of files big. Well, what if you could have a tool that then goes and looks at a skill that you supply to it, and then performs all

1:01kinds of like security analysis on it? And that is essentially what this is, and the reason I particularly like it is that it has Nvidia's name on it. So, you know, it's very easy to put together something and say to Claude, "Oh, you know, what's going on in this skill?" But, if you want someone that could possibly take the blame for their, you know, system not working very well, then Nvidia is a very good name to have for that. So, the fact that they put out a tool that will check a skill, and they're going to stand by it and put their name on it, that's really valuable

1:29to me. So, that would make this tool, you know, better than just some random tool from someone else for doing this task for me. And Peter, you're of the camp of if you get a skill off the internet, before you use it, run this, and don't even use the skill exactly as is. Tell your agent, figure out what I should be using from this, and recreate the skill

1:48essentially, right? Yeah, and this tool basically provides a whole bunch of different scanners for different things to make sure, you know, a skill's not exfiltrating your data out to some other person's back end or whatever. Uh you know, so if you don't know exactly how to go through a skill and look at it yourself and look at all the code that's going on in there, then this could provide you with some, you know, good cover uh for all of these risks that, you

2:11know, are listed here. Okay. I've got a video here that I think is helpful for anyone who wants to really get into this. We'll have a link to this whole report below. Let's go on to the next one. HyperFrames is one that I've seen beautiful YouTube video examples of. People have created these incredible uh files and still I don't use it

2:28because I've got a a team of editors. Is What is this exactly and why should people care? So, to cut cut down to the absolute basics, you can easily create a video just by creating some HTML, some JavaScript, and CSS that performs some kind of animation or moves pictures or videos around in the screen um and turn that into a video. But, the people behind HyperFrames figured out how to do that and also render it at faster than real time. So, you can create code that produces some kind of effect and then render it to an MP4 really, really quickly. And so, this allows agents to, you know, you can tell

3:05an agent, "Make a video for me." and it'll produce the HTML and CSS, and then HyperFrames then renders it into the MP4 file that you need. Um and then on top of that, they sort of added a whole bunch of workflows for doing things like um taking like a talking head video and then putting subtitles or captions along the bottom. Or you can give it like a data set. So, if you've got some sort of data file um that you want it to turn into like motion graphics of like charts and things like that, this will do that as well. So, that is very valuable um

3:34and it's stuff that, you know, you don't really want to be doing by hand cuz, I mean, you know, as any uh video editor will know, it can take hours and hours to do that type of work. This can do the grunt work for producing the videos for you. And what I've seen people do with this is, essentially, they create text that's on a green screen and then um they overlay it on their own video, or they go to HeyGen, and they create a basically an AI avatar of themselves that says stuff, and then they put what they use hyperframes uh to create on top of it. Is that Yeah. Does that sound

4:07right? It's very flexible. So, the fact that it's just working with things like HTML and CSS allows you to do anything that's possible within the browser environment and turn it into a video. So, yes, you can do, you know, almost anything with this. Okay. All right. And it is now in the Codex app really easily, and I've got an

4:25example of somebody who who did this. All right. So, that's hyperframes. Really similar is Remotion. What is Remotion, and how is it different? So, Remotion is kind of the similar idea, but it's based entirely around the React ecosystem, which, you know, people will be familiar with React apps. They power most, well, you know, a large percentage of the web now. They're it's like the most popular front-end

4:50framework. The And so, this kind of popularized that idea, and when the idea of using agents to produce videos took off, Remotion was the first target cuz it was the thing that existed out there, and it basically inspired the hyperframes approach. It does have some downsides, though. Um you know, obviously a big upside, it's been around for years, and people were originally using it um you know, writing code by hand the old-fashioned way um to

5:14produce videos, which is great. It can be driven by agents and AI, but it has some licensing quirks that kind of limit its use. So, I believe if you're a company of up to three people, you can use it completely for free, um but if not, then you have to sort of pay for a license and so on, which, you know, is fine because they've got to

5:35kind of fund their development of this. But, that's one thing that you don't have to worry about with hyperframes. It will just, you know, you anyone can use it. It's completely open source. So, this video right here from Aero explains the difference. for your team, you will have to pay for each render. If you're a hobbyist, it's fine. HyperFrames is a lot That's That's the difference here that license for your team, you will have to pay for each So, no commercial use at any scale. It becomes a real issue when when I'm doing these videos for work. And I guess everything could be considered

6:11commercial. It can be. Um of course, it depends if you're under three people, but it's you know, that's the real barrier on the licensing. But Remotion does have things going for it, which is um you know, its age and maturity. And it has a really good kind of like editor that you can go into. So, once you've produced one of these animations with it, there's this kind of editor built in that you can

6:32move things around and do stuff with. So, there is more maturity around the tooling of this, you know, whereas HyperFrames is more a create HTML, CSS, turn it into a video. Um they're adding things to it, um but Remotion has got, you know, velocity on its side. Okay. Next. Oh, next is uh Zapier. And you all know I love Zapier MCP. Every time there's a new AI tool out there, I seem to give it

6:58access to a bunch of my other tools. This is the way that I get to protect myself. Yes, I want them to have access to read. No, I don't want them to have access to delete, etc. You just get to pick it, and I love how it's just checkboxes. And then at the end of it, you have a simple URL that you give to

7:12your agent, and it has all the power. And if I want to now duplicate every experience that I have with a brand new agent, I can just take that same URL and give it to the new agent. It really works beautifully. It's easy. And if it's great for you, it's even better for your team because they're not as prepared as you are to protect your data

7:30and theirs. So, zapier.com/mcp. Next. Firecrawl. Everybody has this type of reaction to it. Firecrawl, my secret AI tool. This thing just took off. And you weren't super enthused about including this in the list. And the reason I did, by the way, is because it just has this trajectory. I don't think it ever made a single top 10 list of weekly trending repos. And at the same

7:56time, people are talking about it a lot. No, no, no, but I I I do actually really like Firecrawl. Um Okay. I mean, they started as a basically like a web fetcher thing. So, you could just use it to load pages from around the web, and it would do all the proxying and all that type of thing. Um but now they've extended it so much. So, it's got um like search feature in it. It can uh scrape like an entire site, and it can look for links on a site, like you kind of like your own mini Google kind of

8:25spider type thing. Um and they've also added this monitoring feature to it just now, which I've been playing with today, actually. So, you can say to it, um "Look for uh you know, content that relates to my industry, and go and look around these different sites, and do that every 6 hours, for example." And so, you create that monitor, and then that will run then every 6 hours, and it can either send you like an email, or it can ping a webhook that you've got, or you can go and just look at the content on the website. Um and it will basically just do crawling of the web and looking

9:00around for things, but in a kind of a an AI-powered way. So, it's quite intelligent, it's quite good, and it is a paid service, but um they give you like a thousand free credits a month, so it's enough to kind of get started with it and see whether it actually fits in with your workflow. So, no, it's good, and they've really matured, you know, it's one of the sort of the better systems for, you know, crawling web content, or even just fetching individual pages that's, you know, out

9:24there at the moment. So, this came as part of my subscription to Hermes Agent. I forget what it is, like 10, 20 bucks. So, I signed up. I still don't use it because Hermes can just use my browser and start looking around the internet. What's the advantage of using this? Is it that it's now offloaded onto their servers, not

9:43mine? Is there something else? Well, that is one of the benefits cuz obviously, you know, if you're just loading up web pages from your local machine, um you might get kind of rate limited or you might get blocked at some point. There might be some reason why you don't want to do that. Um you might be on the road, for example, and not want to fetch all these large pages locally. So, it can be useful to have someone in the cloud doing that task for you and that's basically what Firecrawl does. Or, you know, if you build an app that is itself hosted in the cloud somewhere, you don't, you

10:14know, it's not going to pipe through your home machine or whatever. You know, you want it to have this kind of third-party service. So, you know, Firecrawl can do that. I see. And you know what I did? I created an app for myself that monitors news stories, that monitors what's hot on YouTube, so that I'm aware of what's going on. And um I can see then baking this in as a way of searching without having to keep

10:34my computer running. All right. Yeah. Firecrawl. Now, this one is kind of a I I find that these are kind of silly. Here's the idea behind this one. I think actually it's easier to show this. What they've discovered is that if you give your agent this type of screenshot, it is cheaper than giving it the instructions by text. And so, what this does is it turns your your instructions into

11:02screenshots. There issues with this, but I think it's just such a creative thing that it keeps blowing up on Twitter. Here's one from Polymarket. Just in, developer claims to have found a way to cut Claude API bills by up to 70% by exploiting image versus text token pricing. And so, boom, it took off, especially around Fable. What do you think of this

11:23one? This is very much an experimental thing. This is for, you know, people that playing around with LLMs and playing around with agents and want to kind of just push the limits a bit and see what they can do. Um but this is kind of an example of that uh whole kind of concept of the picture

11:38is worth a thousand words. You know, people have found I think it was like October last year, October 2025, people kind of discovered this idea that if you you know, images don't use as many tokens to represent what's within the image as the equivalent text would. So, essentially it's a form of optical compression. Um and if you push text in via an image, it uses fewer tokens. The problem is it's not entirely accurate, so there are some benchmarks on the PX um pipe repo that kind of explain sort of just how well it works and you know, how well

12:15it doesn't work sometimes. Mhm. But yeah, it's an interesting idea. It's not something I would just jam I wouldn't jam this into my, you know, workflow at all cuz the fact of saving a few thousand tokens here and there is not relevant for me. Um especially if there is a downside to the accuracy. But as a concept, it's very interesting and it just shows you how far the LLMs have come that they can even do this. You know, the fact that they can take an image that's packed full of text and reasonably faithfully reproduce it internally is just kind of cool. So, yeah, it gets my cool rating but

12:49practical rating not so much. I did find this for controversy. Dense image content can be misread and when it is, the model may quietly invent a wrong value rather than flag it. Yes. And that's the thing. be useful for math. Like things like, you know, mathematical equations and things. If they can jam that in, then of course you are actually using more visuals in there. So, it can be better for reading certain types of thing but yeah, I wouldn't just push this in front of my approach. It's there's too much of a

13:19risk. I think these are really good for getting a lot of views on Reddit and on X. Yes. [laughter] Okay, Baidu unlimited OCR. Baidu's OCR model can read long documents in one shot. Here's the tweet that shows what it looks like. I'm going to show it here on my screen in a moment. Um, this looks magical. It's quick. Um, here is the growth chart on it. Um, it took off and for some reason it never made our our weekly chart. I do have

13:49this example from hugging face. And I then took the exact thing that we just saw a moment ago, the dense collection of text. Let's put it in here. I'll hit start and let's see how it does. And here is one of the issues that we should bring up with it. It's waiting for a GPU slot, which means I can't run

14:09this on my Mac. No. It is literally Nvidia only. That is, you know, the the key thing about this. It's all CUDA and all that type of thing behind the hood. Um, I did actually run this today just to see how well it would work. Yes, and I got about 90 tokens a second

14:27output on my 3090 Ti that I've got. Which [snorts] is pretty good. Um, and the thing the interesting thing about this is it's based on work that was done by Deep Seat with Deep Deep Seat OCR. But they changed, and this is kind of a technical term, they basically changed the way it works to using a sliding window of attention. So, when a a document gets larger and larger, some systems can become slower and slower because the amount of attention needed is the basically the way it scales up is very complex and just increases

14:57quadratically, I believe it is. Uh, this uses what's called a sliding window, so you kind of have a more constant amount of attention, but that basically means for real life use cases, you use less VRAM and more consistent speed over a run. So, if you've got a long PDF, it will run at that 90 tokens a second for, you know, many, many pages rather than getting slower and slower as

15:17it goes. Um, and so it did in my example that I ran earlier today. So, if you've got a lot of PDFs that are very complex, maybe you want to read them in and especially if they've got like, you know, they're not just plain text but they've got a lot of like layout elements and things like that, this actually works quite well. But, you do

15:35need an Nvidia card. You know what? When I hear you talk about this, I'm just amazed at how much you know and how the level of rigor that you bring to these repos. Can you tell people your background and why you're able to discuss this so intelligently? I'm just a nerd, basically. [laughter] I'm just a nerd who's been playing around with computers for 30 or 40 years. So, And a publisher [snorts] Yeah, exactly. When I when I see stuff like this, it's just I just have to, you know, play around with it and I'm just amazed at, you know, what I see a lot of the time. It's just moving a

16:06lot faster than you know, I could have ever imagined. So, yes, I just play a lot. I just have a lot, you know, a lot of time for this type of thing. And of course, this is how I discover how to do things. So, you know, I run a bunch of newsletters. I'll just, you know, pitch one which is

16:19javascriptweekly.com. So, if people want to, you know, keep up to date with everything that's going on in javascript, that is the place to go. So, I just basically keep up with all of these different things. So, I've got the broad knowledge of what's going on out there and it just feeds back into my work. So, yeah, and it's fun. So, Yeah, super fun. All right, and we'll of course have links to that below. Let's

16:39take a look, by the way, at the result. It stopped. It says, "Text 0099 etc." It says, "The following is system prompt tool documentation renders it as an image as an image for token efficiency. OCR should run in the context of the image. The output is an empty string. Typeface, the output is an empty string." So, that was the part that convinced this did a great job for you because I mean, the image that you supplied is extremely dense and extremely hard to read. I mean, even a person could go through this step by step, but really this unlimited OCR thing is better for, your PDFs that actually have

17:19proper size text and I see. I ran it and it worked totally fine. So, on this this was really a challenge. Okay. That's why I wanted to do it. I wanted to see how it worked here, but that makes sense then. All right. And I've got a video here explaining it in more detail. And one of the things that he did here, um Little Coder also took a handwritten note and put it on and it was able to OCR the handwriting, too, which I thought was interesting. And of course,

17:43we'll link to it. Okay. An open-source AI coding agent that lives in your terminal and isn't tied to any one model provider. This is what people have been looking for. This one has actually gone um gone around in my circles a lot. It's not new. It hasn't had this major breakout and that's why I

18:01wanted to include it here. What do you think of this one? Um I like it. I've used it quite a bit. It's not my main um agentic coding kind of harness by any means, but it is really, really slick. And I know a lot of top developers use it. Um I believe the the guy who created Ruby on Rails, I think this is his kind of uh

18:23you know, tool of choice in this regard. Uh it's JavaScript based, so kind of relates to exactly what my sort of main, you know, work is. But the thing that is the the main selling point of it is that you literally can give it the API key to whatever it is you signed up to, whether that's, you know, the Anthropic stuff or OpenAI or Deep Seek or whatever, you know, system you're on and it will pretty much deal with it. Like, there are you know, support for hundreds of different models in this. And you can switch between them as well, which is kind of interesting. So, you you can you

18:52can add things in to say, you know, run my agent on this cheaper model and, you know, run my other thing on the big model and so on. So, I just really like it. Um and it's a little bit bigger. There's another tool that's similar called Pi, which a lot of people may have heard of, um which is at pi.dev, um which is a much more lightweight version of this. But um yeah, if you want something that's like a Claude code experience, but completely flexible, you can keep up with the costs, you can put your own models in, this is for you. Um and it's completely

19:19open source and everything, so very cool. But it doesn't work on the subscriptions the way that I use Claude. That's a contro- that's a controversial point. Obviously, people have tried to use it with that and they kind of notice that you're doing it and stuff. It's the it's the same problem with Open Claude

19:34all over again. Um but there are subscriptions that will work with it. I think like GLM. Um I think the GLM stuff will work with it. I think, you know, the Deep Seek subscription works with it. So, there are options. But yeah, probably not Claude. Okay. I'm I think also that um Chat GPT does work with it. Yes, I haven't I can't keep up with this all the time, but yes, there are different people that are always kind of sucking up to developers and allow people to do things that other models

20:02don't do, so. What's the advantage of to open code versus Actually, I guess open code is the big one, right? This is the one that I hear about most. Is there anything that makes it especially great that gets so many people talking about it? I think the interface is just really well thought out like in some ways and I couldn't enumerate all the different ways, but in some ways it feels nicer than Claude code. Just kind of just feels really slick and I don't know, I've never run into any bugs with it or any weird oddities, so yeah, it's good. I like it. And the fact

20:34it's completely open source, unlike Claude code now, Yeah. yeah, makes it a natural choice for people that are really strongly into that, you know, when they're running on Linux and they want everything to be open source, this is a great way of being able to get that Claude code experience without using Claude code.

20:50Okay. Next, number eight for the week is Agent Cookie. I've seen this a lot because a lot of my friends will run multiple Macs and what they do is they'll have one Mac for uh this type of agent, one Mac for this type of work, and then of course, they have the one that they have on their desktop. And the problem is that when you're on your computer, the browser that the agent uses basically has the same access that you do. It's logged into Twitter. You don't have to give it a username and password and then get past whatever Twitter's anti-robot systems are. But once you go

21:21to the second machine, now you have to go back in and log in and it becomes a real pain in the butt. And so this is from Matt Banhorn, someone who I tried to get on here before and he thankfully did an interview with me. It'll be on the channel. Um and he said, "I'm going to solve it." And he's got his fans. What do you think

21:40of this? You've actually summarized it really well. There's not a lot more I can add to that, to be fair. You know, it is a common technique to reuse your browser cookies to allow agents to access things as if they are you. Um and then of course the problem is is that if you've got that Mac mini that you bought for Open Claw, it can't do that all of a sudden. You can copy them across manually, but this basically provides an automated way of doing it. Um but it does require that you use Tailscale to kind of keep your machines connected. And I mean, I don't

22:09know if that's a tool that you've ever covered before. Um but Tailscale basically allows all of your different machines to act as if they're on the same network. Now, they might well be if they're all in your house, but let's say, you know, you've got different machines and you're out on the road. You've got a Mac mini sat at home doing

22:26stuff. It's got your Open Claw on it. You're out on the road with your MacBook Pro or whatever. Um Tailscale allows your MacBook Pro out on the road to look as if it's on your local network. So it allows that syncing to occur no matter where you are in the world. And that is one of the reasons that people use Tailscale and that's what this project also requires to kind of have that virtual kind of like local network running all the time for copying

22:49the cookies across. Yeah, I feel like that's something that we should cover at some point here. Again, that's one of those tools that everyone with an agent is talking about with a Hermes or Open Claw agent. All right. Sounds like issue. Now let's get to something else here. Caveman basically helps you save on token costs by reducing the way that you interact with your agent to Caveman speak. I think that this from their uh readme file really explains it. Instead of all this text, the reason your React component is rendering is likely because you're creating a new etc. You go into new object ref each render inline object

23:27prop equals new ref. You get it, you know? And I I think they have other examples of levels of simplicity in language that reduce token cost. Um I I think this is another novelty thing designed to get attention, but really not super effective. Though it is, from what I'm seeing, much more effective than the render your prompts as images

23:51repo that we saw earlier. Yeah, I mean, I wrote some notes for this and basically I wrote me cut tokens by talk like Caveman. Like that is the That's the Caveman way to describe this project. Um it's actually based upon There's kind of been this long-running kind of meme in the developer community called the Grug brain developer. And people basically explain development stuff in this context of boiling it down to how a Caveman might stereotypically

24:18talk. Um It It's interesting. Again, as you say, it's a novelty. But if you want the agent to talk to you in this kind of style where it just cuts all that nonsense, all that kind of cloudy kind of Oh, I quietly did this and I'm delving into this. Maybe you would like this kind of communication style. Um And one of the cool things that they have added and it's mentioned in their repo is they've they've created their own kind of like Claude coder like called Caveman code, which they claim will actually get you the most token savings and it it will actually try and make the

24:51agent think in this kind of Caveman speak as well. Now, I dread to imagine what the results of that would be. I hadn't tried it out, but uh it's fun to see him experiment with this stuff. So, definitely a novelty thing. But, if you like that type of output, it's worth a try. So, um yeah, it's not completely

25:06useless by any means. I see. You're saying in addition to whatever token savings you get, you also get a a potentially better user experience because a lot of models are just overly verbose, the cuteness is annoying, and this eliminates it. I've also heard that the that the cuteness, verbosity, ends up confusing the agent. And so, you end up with potential mistakes and problems because of it. Is

25:30that right? I have no data on that whatsoever. Um I don't The only thing it did remind me of though was we in a previous show we talked about a tool called Pony Tail, Yeah. which I don't know why it reminded me of that. I think it's more the fact that it you're trying to get the agents to communicate in a different way with you and act and think in a different way to how they are trained. And that's just an interesting concept because sometimes it means that the output comes out very

25:59different to what the default would be. And whether that's better or not for your projects, you have literally have to try them and find out. You know, a lot of people have found success with things like Pony Tail. Um again, maybe it's something that could work for you, this Caveman thing, but you it's a one of these things you have

26:14to try it to see whether it does or not. You probably just get sick of it in like 10 minutes. I suspect that's what would happen with me. And you know what, though? I shouldn't be putting these down. I find that sometimes things that start out as little silly games or little silly experiments take off in an interesting way, and let the let the creativity flow. I'm

26:33happy with that. I have a perfect I mean, I have a perfect example of that. So, Chain of Thought, which is now, you know, a common concept in LLMs, this whole think thing where it, you know, you give it a prompt and it says, "Oh, let me think about this and let me work through this." And then it gives you the result. That actually came out of similar experimentation back in the GPT-3 days, you know, when models literally would reply immediately to what you put in, people were adding things to the prompt

26:57saying, "Don't reply straight away. Think about Think through your process. Think through the result, and then give me the answer." And people were manually doing that. And then, hang on, all the different LM providers were like, "Hang on, this actually works really well." Um [snorts] and then they baked it in as an actual behavior of the model. So, these

27:15things are not useless by any means. This is all worthwhile experimentation. And you never know what's going to end up, you know, in your models at the end of the day. Maybe your model will start thinking like a caveman before it gives you the final result. So, yeah, it's cool in that regard. Way to go, Julius. I'm glad we got to feature it. Final one, browser use. Hand an AI a real web browser and tell it to

27:35just do the task. You were kind of meh on this one. Uh yeah, so I I thought I would go and try and use this one as well. So, it comes in kind of like two variants, uh which is that you can run like your own kind of self-hosted browser, like headless browser, and then have this kind of drive it to do various things, like you know, like in Codex, I think Claude as well, they've kind of got this browser use feature now, where it can you can tell it, "Go to a website and fill in these things on the site and click on these

28:07buttons and perform various actions." This is more like a self-hosted kind of version of doing that. Like, if you don't want to be tied into the OpenAI or the Anthropic way of doing it. The other thing that they have, which I've been playing with, is they have a hosted service, which is really what this is intended to promote, where they will host the browsers in the cloud for you. They charge you like 2 cents an hour, I think it is, to run them. So, if you want it to perform a task that, you know, literally takes a minute or two to do, you're paying like fractions of a cent, um if you go

28:42through their cloud service. Now, if you do it completely locally and self-hosted it, obviously costs nothing. But even using their service, which I was using earlier today, actually worked really well to, you know, just give it a task and it would spin up for, you know, 10 seconds, perform the task, and then be

28:55done. Um you know, and they have like a free tier as well. So, it's it's worth trying out. I quite liked it in the end. Okay. Yeah, it um it had gotten a lot of attention. Again, not enough for it to have made one of our top 10 lists for the week, but I thought it was really interesting. Again, we've got another video here for people who want to see it. And I've been adding these links to the bottom to sections that I think are helpful. Truthfully, this is done with AI, but I spend so much time getting my my skill to pick the right spots, and

29:24it's doing a much better job at it. So, lately I haven't even needed to edit it. All right, it is in there for you all. And as always, oh, no, we have 11. How many did I [laughter] put up this week? You got a bonus. 11. Your own private Google Photos running on your server. Image is a self-hosted photo and video backup app you run on your own hardware. I didn't I stopped at 10. I just assumed we were done. Is this one of yours or did this

29:50come from my research? This was one of mine. Um they added uh there's a brand new major version just came out. This is actually quite a mature project. It's been around for a while. Um I mean, as you can see there, it's been around for a few years. If you want something that is a bit like the Apple Photos experience, but you don't want Apple to be hosting that for you, you can run Image on your own server. You can install the app on your

30:12Android and, you know, um iPhone. Uh and then sync your photos and videos up to your own personal server instead. Uh and I know this is something you've spoken about before, that you were looking into, you know, hosting your own files and stuff like that, without relying on things like Dropbox. This is the same idea, but for photos and videos. Um you know, and it's very mature, works very well. So, it's

30:34something that people might want to do. All right. I had no idea. All right, I I had no idea that this was in there and that I'd missed that I'd missed it. Um, let's see if there's another. There is another. Graphify. Uh, this one actually is one I think that I had picked out and I thought we'd reduced it to 10. I'm glad to see that we've got more here. You're going to turn a folder

30:54everyone. Yeah, [laughter] I you know what? I usually go beyond the 10. I will do a top 10 list and then I'll find two or three that I've heard people talk about, but this week is all that. Only things that I've heard people talk about that did not make the top 10 list uh, the way I thought they should. So, Graphify you're going to turn a folder of code into queryable knowledge graph. We've

31:13seen this actually. [snorts] Um, this is an issue that's come up a lot. How do you make your data like easier basically put an index like you would on a book on the data that you have so that it's more searchable, that it's faster to get access to. What do you think of Graphify? So, I mean just to really

31:33hammer home the concept of what this is. If you go into a code base with something like ClawCode or Code Atlas and you say, uh, you know, change something within this app and that change involves lots of different parts of the app. What those agents have to do is they have to read the source code files, figure out what relates to what, and then work out how they can make the change surgically by changing, oh, I need to change this piece of the app and this piece of the app and this piece of the app. Maybe I need to look at these docs that are within the app. But it has to kind of

32:02figure that all of that out as it goes. Something like Graphify, what it does is it will go through and analyze a project first and figure out okay, this documentation file is for this piece and this piece and it will figure out which parts of an app are related to each other and then kind of put this together into a structure that is more kind of well, just a structure that is naviga- you know, an agent can navigate through rather than having to try to figure out everything from first, you know, principles every single time you want to make a change. So, this is one I've not tried out and I've not really

32:33dug into a lot, but the the concept is sound. Um, and it clearly is very popular. All right. Um, so now I know this one is the last one and I've got all the information here, this whole report below. If you like it, of course, like, subscribe, comment, do the whole thing, but more importantly, Peter, we've got a link to you below. I urge people to go and connect with Peter directly on his channel, on his newsletter. And I will also say that now that this is over, Peter, the number one video on this channel is the one that I recorded with you. It's got over 100,000 views, which

33:07we never had before or since, unfortunately. I'm going to link to it right here and you all can see the most popular set of repos with the most popular guest on the show. Click over and I'll see you there. Thank you very much.