Open Minis Turns iOS into an Agent Playground

Hello and welcome to another episode of AppStories. Today's episode is brought to you by InnoReader, Claude, and Scribe. I'm John Voorhees and with me is Federico Viticci. Hey Federico. Hello John, how are you? I'm doing alright, I'm doing alright. I'm excited for today's episode because I think we have a lot of really interesting topics. We've talked about a bunch of apps, you and I, over the last couple of weeks. So we've been trying things, you wrote a big story about Open Minis, we have another one to talk about in the pro show. But before we get to the apps, I really want to talk about a couple of things that I wrote about this week. And I know you talked about it a little bit with Mike on Connected, but I kind of wanted to expand on the AppStory Chaos story that I wrote. And I thought it was really interesting to hear your reaction to it because the point of the story was there was a New York Times article that basically was describing how we're heading into the second half of 2026 with a projected app submission that is going to more than double last year's and be an all-time record. Yet at the same time, downloads by customers is only increasing by 2% or 3% year over year. And that's obviously a problem, potentially. And you and I have experienced this very firsthand with the number of emails we get, which sometimes I look at my email inbox and I just despair because there's so much there. And what I wanted to talk to you about in a little bit, I know you have a strategy and you mentioned to Mike how like a Gmail address is kind of a tell. The one that I heard you mention that to me I hadn't really even thought of as a tell, but I thought was really weird that I was seeing it everywhere all of a sudden, is that the app is secure and private by design. Yeah, that always happens. That line is everywhere. And it's even where it makes no sense, where you look at the app and you say, well, yeah, of course it's secure by design. There's no server component to this app at all. All it is is like, I don't know, a list of chores that is local on the, you know. And at first I had, I was thinking, well, this is just, Apple's been really good at its marketing about privacy and a lot of third developers, third party developers are picking up on that. But I think you're absolutely right that it's part of the typical skills that people are using to build these apps and come up with the notes for the release notes for the app store and that sort of thing. And when I wrote that, I ended up the piece kind of optimistic in that I do feel like these tools are very good for new developers. That's going to bring in new ideas and there are going to be hidden gems, including like open minis, which we'll talk about in a little bit. But at the same time, there's a flood of stuff now that feels hollow and soulless to me is how I would put it in a way that really it makes it very hard. And it's very hard to differentiate things the way it used to be because it used to be that creating an app was a hard thing to do. Yes. Right. And so when you and I got a pitch from somebody, we could tell right away whether it was somebody who was like, well, they may be a first time developer, but this is a person who obviously put in the time and learned how to code, learned about product, learned about design, did all the things you had to do. Or sometimes they hadn't and it was just not a very good app. And it was very easy to kind of make that call right off the top. Yeah. Today, when we get those emails, all of them have the veneer of a, they have a veneer of being a good app in a way in that, you know what I mean? There's like, there's this sheen of this is a legitimate looking app. But when you dig into it, you realize it's really pretty. It's cookie cutter. It's like a form app or something almost. Yeah. Yeah. This is what happens when you democratize the ability to, for anybody to create software, which is a great thing if you ask me that just anybody can put together an app and put it on the app store. But that also means that anybody can create an app and put it on the app store. And it used to be that the act of building and releasing software was very intentional and very exclusive in a way, right? You had to study programming and computer science and be a developer by trade. And now just even people who are not developers by trade can do it, right? Right. And in a way, it's not too dissimilar from the abundance of content that we're seeing in other industries, whether it's music or to an extent, maybe video games or blogging, right? Or creating videos on YouTube, anybody can do it. And I think if those other industries are of any indication, I think at some point it'll balance itself out in that the people who really care and the people who really truly release quality products, they will bubble up to the surface. It just means that it'll be more difficult for those gems to be discovered and for people like us who need to discover those gems to navigate this sea of just thousands of new apps on a weekly basis. But I don't despair too much because it's becoming quite easy to tell, right? What's a cookie cutter project and what isn't. So it's changing the landscape of the industry. It's changing the numbers. It's changing the app store. What's not changing is that you can tell when you're dealing with a developer who really thinks about their product with a product perspective instead of just saying, yeah, Codex, put it on the app store. Yeah. In a way, I think in that sense, the email has become more important. Like what somebody says in the email that becomes the tell as to whether they've really put the thought into the thing or not. Yeah. I mean, I think for me, the thing that I struggle with, I'm not again, like I said, I am ultimately optimistic and I think it is good for more people to be able to create apps if they want to create apps. The thing that I have found so difficult is just dealing with the volume because you have to kind of look at an app to figure out if it's any good or not. And there are some easy early tells in like the email or something. But I am finding that the volume, even if you only spend a few minutes on each one, it's almost overwhelming at this point. Yeah. Yeah. Well, I thought it was an interesting article, especially based on the data from sensor tower, which is not always super accurate. But I think anecdotally speaking, it is accurate. Like it is a pretty good reflection of what's happening. Yes. I always take those numbers since they obviously don't have access to Apple's data with a grain of salt. On the other hand, I do find them to be, as they would say in the corporate world, directionally correct. Yes. In that over a long period of time, the trends of up versus down and magnitude of downloads and submissions and things may not be exactly right. But they are relatively correct at, you know, year to year and that kind of thing. This episode of App Stories is brought to you by InnoReader. InnoReader helps you build your own personalized news feed by following websites, newsletters, YouTube channels, podcasts, social media feeds, and more in one place. Instead of relying on social media algorithms or jumping between apps all day, InnoReader lets you follow the sources you trust and get their updates directly. For anyone trying to keep up with a lot of information, that can make a really big difference. You can create custom monitoring feeds that track keywords, brands, topics, companies, competitors, products, or browse industry trends across the web. InnoReader also gives you tools for building a personal knowledge archive. Save articles for later, highlight key information and notes, and search through everything later when you need to come back to it. And with InnoReader's intelligence features, you can summarize articles, extract key insights, and make sense of larger volumes of content in less time. There are powerful filtering and automation tools, too, so incoming content can be organized automatically with the most important updates surfaced while distractions are pushed out of the way. InnoReader is available on the web, iPhone, iPad, and Android with offline reading support on mobile. You can try InnoReader for free or explore everything the pro plan has to offer with a 14-day free trial and no credit card required. New pro subscribers can get 30% off their first year. Visit InnoReader.com to learn more and get started. That's I-N-O-R-E-A-D-E-R dot com to learn more about InnoReader. Our thanks to InnoReader for their support of AppStories. All right. What else do you want to talk about? I wanted to talk about TestFlight and what's going on because you guys talked about this, too. And what I, you know, I discovered a couple of days ago as we record this that TestFlight got an update that added search, which I was really glad because I had written an article a couple of years ago with a long list of things I wanted to see come to TestFlight. And the top one was search. And that's simply because, you know, the lists were really getting really long. But when I opened up the app, both on the Mac and on my iPhone, I found that not only was search added, but now the list, instead of being reverse chronological order with only active apps visible at the top of the list and inactive ones in a section at the very bottom of the app, everything was alphabetical. And betas that I hadn't seen in the app for years had suddenly appeared. It was literally every single beta I have ever tested. And now you told everybody on Connected that I have 800 betas. I do not have 800 active betas. How many do you have? I don't know how many active ones I have because now they're all dispersed between all these other ones. That's 800 that I've probably ever tested in 11 years at Mac Stories, okay? So what I did is I took a video of me scrolling it, then I gave it to Codex to freeze frame it and count the number of apps. And it came up with a number that was like 777. And so as I scroll through that, it's rare for there to be more than one to three active betas on any screen on my iPhone, which is rough because if you're looking for something, it's very difficult to scroll a list that has 800 things in it. And while I know that you and I are unique, we're not the only ones who review apps that have a lot of betas. We have special snowflakes after all. Yes. Yeah. I mean, there are other people. There's, you know, people like Jason Snell, the folks over at 9to5Mac and MacVermers and everywhere. But I don't understand why they would change the sorting without also giving users options on other ways to sort it. So reverse chronological order, I actually liked a lot because it meant the newest betas were on the top. And those were the ones that you wanted to check out first usually. And something that was like at, you know, going to expire in three or four days was probably already on the app store and you wrote about it or you had decided not to. So it was usually you could get, you know, there was a good density of interesting new betas at the top of the list. Now they could be anywhere in the list. I've never really understood why inactive betas and expired betas were even included in TestFlight. I don't know what purpose those serve, even if you're not an app reviewer, because the developer has taken them away and there's no way to get them back from that list. It's just a leftover. Like, just maybe put those in your account history or something. Like, if it's necessary to see that at some point you tested that, maybe create an archive page that's separate from the currently being tested betas. They shouldn't load in the main view because I will tell you that... It also slows down the main page. Oh, it really does. And I have had, I mean, back in 2023, I actually wrote about it because it was so bad that I could no longer get TestFlight to load betas and it was completely broken for me. And they fixed it, but they haven't completely fixed it in that it's still sluggish if you have a lot of betas. And the problem isn't just active betas. I know I got a lot of grief last time I wrote about this because people are like, oh, you have too many betas. How could you possibly ever test all those apps? And it's like, you know what? I don't. I test the ones that I find interesting that I'm going to write about and the other ones are just there. Plus, the slowdown isn't just the active betas. It's the fact that all these old betas are in there and there's no way to remove them from the list except one at a time where you open it up, scroll to the bottom, tap a button, wait for it to disappear. There's no bulk editing. There's nothing like that. So this is a big problem. And there's a twist. First, because as soon as I wrote about this, I heard from a couple people on Mastodon and I heard from one of our writers, Devin, and I heard from a couple of club members in Discord that they weren't seeing the same thing. And it turns out one club member told me they were seeing it on some of their devices, but not other devices that they have. And so I'm not sure if this is a bug or if this is an A-B test or what. But first of all, if it's an A-B test, I'm here to tell you, Apple, this is a bad idea. Don't do it. And if it's a bug, boy, I hope it gets fixed soon. I have no idea what they're thinking. I have a hard time imagining how a bug takes a list that had two separate defined sections with their own headers and a table view and combines them into one list and resorts them as alphabetical. Do you know what I mean? And it's like a design thing that seems intentional, not just a mix-up of how the code works based on what I know about iOS apps. Plus, it's on all platforms too. So I don't know. I really don't get it. If it's intentional, that's a bad intention. Like, let me tell you, if it's intentional, like it's a bad idea. You screwed up. It's bad. There's no way around it. I don't want to try and excuse a bad design when I see it. Like, it doesn't make any sense to mix multiple betas of different kinds in the same list with no filters, no sorting, and just a search box at the bottom. It doesn't make any sense. Right. And I will say that, you know, people might say, well, why don't you just automatically update the betas? And there are a couple of reasons. One is on my iPhone, it hasn't worked for me for years. I haven't been able to automatically update betas even if I wanted to for years. So I have to do it manually. But even if I had that ability to do that, sometimes I actually like not updating them automatically because I want to have, I want to go into TestFlight, see what the developer says about what's changing before I install it. Because sometimes I already own the app and it's just like a small bug fix update that I don't really want to deal with at the time. And sometimes it's like a major release and I want to know if it's a major release versus like some little stuff that I don't really care about and won't write about. So, yeah, it's while we have hopefully Apple's ear, I would say don't stop at fixing what you broke. Get to the filtering and sorting. Make the thing way more performant and give me the option just to get rid of all those old betas because I'm not going to go through and delete like four or five hundred one at a time. We're not. And I really think they should just be displayed in a separate page. Just like on the App Store, you can go into your purchase history. If it's really necessary, and I continue to doubt that it's necessary to show those expired things, but if for whatever legal reason it is necessary to keep displaying them, put them in a sub page in your account, like testing history or beta archive, like whatever you want to call it. Just put them there. In over 10 years, I have never once looked at expired betas. Why would I? If it's expired. And it's not like the App Store where technically speaking, I could go into my purchase history and re-download something from the App Store, right? TestFlight is not the App Store. I'm not going into an expired beta and saying, ah, let me download this expired beta. It's expired. It's gone. That's totally up to the developer. And developers sometimes will do like, they'll redo their betas as almost as though there are different SKUs. And so sometimes I'll scroll through that list and I'll see one developer with one app and there's seven or eight entries because of the way they set up their test flight betas. So even that causes even more noise inside the list. Yeah, man. I don't know. I don't know. We'll see. We'll see. This episode of App Stories is brought to you by Claude from Anthropic. Claude is the AI for minds that don't stop at good enough. It's a collaborator that actually understands your entire workflow and thinks with you. Whether you're debugging code at midnight or strategizing your next business move, Claude extends your thinking to tackle the problems that matter. I've been using Claude and Claude code a lot. And I just recently tried the new integration with the iOS simulator. It actually works with the simulators for iPadOS and VisionOS and other things too. And it's really fantastic because you can be working in Claude code in an app that you're building and you can see what's going on in the simulator. And so can Claude code, making it much easier to debug issues and work on the design. If you're a developer spending half your day on tasks that you wish you could hand off, Claude code runs in your terminal, reads your code base, and can take on things like writing tests, refactoring, and debugging without you hand-holding it through every step. You can forget about basic web searches. Claude is excellent at deep research. That means you can dig deeper with extensive sources, proper citations, and do it all in a fraction of the time. Recently for AppStories, I did just that. I was researching the terminal app on iOS called Ish. And I wanted to understand better how it worked and why it was valid under the AppStore review guidelines. And Claude was perfect for that, finding old blog posts by the developers as well as news accounts about the rejection of the app and then its allowance into the AppStore. It was perfect. Also, Cowork can create polished documents, spreadsheets, and working formulas and presentations directly in your file system. Not rough drafts that need heavy editing. I have Cowork helping with this kind of stuff all the time. It can access my email through Sparkmail and then create summaries of the most important emails it finds and drop them into Notion with links back to the emails so I can read through everything, get a nice summary, go through it quickly, and then one at a time go back to Sparkmail via the links and respond to everybody. It's a great tool for doing that sort of work. For problems worth solving, get started with Claude today at claude.ai.appstories. That's C-L-A-U-D-E dot A-I slash appstories. And check out Claude Pro, which includes access to all of the features mentioned in today's episode. Claude.ai slash appstories. Our thanks to Claude for their support of AppStories. All right, Federico, let's switch to something that's a little more exciting and a little less negative. Let's talk about Open Minis because this to me is a remarkable app that I've been spending some time researching the background of how it is even possible that it's doing what it's doing. And I think it all goes back to the terminal app that it forked and that it's using under the hood to make this whole thing work. But I want you to – you wrote the article. I want you to kind of give people the broad overview of what it does and what it is, and then we can kind of dig into some of the nerdier details. Yeah, right. So as I covered on Mac Stories and also explained on Connected, the basics of the app is that this is an app for iPhone and iPad. You can also install it on your Mac in iPad compatibility mode. And it's a chatbot app that lets you have conversations with different models from different AI providers. You can sign in with different providers like ChatGPT, Cloud, XAI, Open Router, just plain API key. Any OpenAI compatible API endpoint will work, and you can just have conversations with chatbots. But that's not what makes it different because there are thousands of apps like this. What makes it different is that this is a first-class iOS citizen. This app was created by a developer who has fully embraced iOS frameworks and iOS integrations. The app itself is not just a regular chatbot. It's more of an agent in the sense that it has its own workspace with a browser and a Linux terminal. And all of this is contained in a literal workspace that is a folder in the files app. The agent is capable of saving memories as markdown files and skills. And like OpenClaw, it has a sole document, a sole.md. All of these documents can be inspected and edited in the files app. And not just that. So imagine it's like a sandboxed OpenClaw built exclusively for Apple users on iOS and iPadOS. And the developer went a step beyond. It turned a bunch of iOS integrations into CLIs that the agent can use. And I'm not just talking about the usual reminders. You've seen reminders in Cloud for iOS. There is reminders in minis. But there's also HomeKit, HealthKit, Bluetooth, Photos, Maps. There's WebKit. There's, I don't know, dozens of Apple. Oh, there's music. There's music. It's a big one. Media playback, notifications. I can keep going, essentially. Imagine all of the frameworks, all of the APIs that Apple makes available to third-party developers who want to create apps for iPhone and iPad. This developer turned those frameworks, bridged those frameworks to the terminal that the agent can use. And so this agent in OpenMinis, which can be powered by ChatGPT or Cloud or whatever model you prefer, is able to use all of those system hooks on the iPhone and iPad, resulting in an experience that is literally unlike anything else on iOS right now. Right, right. And, you know, Photos is another one that comes to mind that I think is big. Because the thing is, is that you're right. Things like ChatGPT and Cloud have integrations for things like Calendar and Reminders. But those, that's about the... ChatGPT doesn't even have Reminders, I think. Oh, it doesn't? That's just Cloud? I think it's just Cloud and Perplexity. Yeah, and some of them have access to the Finder on the Mac. I mean, they can do that too. But that's really about as far as anybody has taken it. And, you know, Perplexity, I think, has done some of the best work in this area. But, and I'm not entirely sure. I mean, I think most of them are doing it via MCP or some kind of modified MCP type of setup. But nobody's, none of these big companies have taken it a step further and tried to integrate any of the other APIs that are out there. I mean, the thing about like Reminders and Calendar, I mean, they're basically both, they're both based on EventKit. So like EventKit has been like the center of attention for some of these companies like Anthropic. But they haven't really, no one's really taken it much beyond that at all. Yeah. And Minis is an example. I think it's so fascinating because, first of all, I have not been able to find any other even remotely similar example. It is genuinely a novel idea to take iOS APIs and make them command line interfaces for an agent. I've been able to do things like, since it's an agent, right? I've been using this with GPT Sol. I've been able to do things like, I want you to check all of my temperature and humidity sensors in my apartment and build me a visualization of the temperature and humidity across the house and organize it by room. And imagine combining the agentic features that we've seen in Codex and Cloud Code with something that truly had access to your iPhone, right? Much more so than Siri AI, for example, because Siri AI has access to those features, but it cannot think and act on data like an agent does. Right. And to take it, I think your example is a really good one. And to take it like a step further, the kinds of things you could do with that is, yes, you could just get the data organized in like an outline in text on inside of Open Minis. But you could also have it create a webpage that creates like a floor plan of your apartment with different color coding for like the temperature or humidity. You could do that. You could also set up a series of notifications, like for when it's too hot or too humid in a particular room that would then come back to you. And you can integrate that all with personal automations because shortcuts is in the mix as well. You can trigger Open Minis sessions with shortcuts. So I have some scheduled automations that start a chat, a thread in Open Minis and do something in Open Minis. And I get a notification when it's done. And there's so much you can do. I mean, I can get into some of the nerdy details. And I should also mention that I'm on the beta of Open Minis. So some of the functionalities are upcoming. But for example, first of all, since it's powered by a Linux terminal, in fact, it's a fork of ISH. It's a popular terminal app for iOS users. Which has an interesting history, I learned. Oh, yeah? Yeah. It was originally, it was released and then it was rejected. And then Apple unilaterally let it back on the store. And it is kind of, ISH is kind of what you, it's day one, day zero of really this whole lineage of having terminal apps available inside of iOS. And game emulation because they're using a similar related technique of doing interpreted code instead of running code. I mean, that's the thing is like when you hear about an app like this, you might think, oh, I thought you couldn't have, you couldn't have an app on an iPhone that executes code. How can you have a terminal that has like a package manager that can download something like FFmpeg and, you know, and encode a video for you? How can you do that? It's because instead of compiling and running machine code on the fly, which is another reason why like JIT is not allowed on iOS. Instead of using, doing that compilation on the iPhone, it's interpreted code, which as I understand it, and I'm sure someone will correct me. But the instruction set for ARM has already been compiled in advance so that when the app is submitted, all the code that will run the terminal app like ISH has already been run through Apple's process of app review and approved before anybody runs anything. And it's a finite set of instructions that can be run. Which is one of the important security things. Because I think we do throughout this have to kind of think about the security because it's, I know that a lot of people would be very hesitant to do things like running terminal apps on their iPhone. Because you hear terminal app on the iPhone and you're thinking, oh no, what if you sudo give it access to your phone? No, like the terminal is a full-blown terminal. You can install Node. You can install FFmpeg. You can install the YouTube DLP. You can install any CLI from the web. But you're still running inside that sandbox. The virtual file system that Minis exposes, right, via this terminal is limited to slash var slash minis. Like it's not reaching into the entire iOS file system. It is its own little operating system almost that is confined to the application sandbox of Open Minis on your device. I think the one place where there are security concerns with something like Open Minis is with its web browser potentially. Potentially? I mean, obviously. And prompt injection, right? Whenever you're dealing with an agent and the agent has access to a built-in browser, you're always dealing with prompt injection attacks. Absolutely, absolutely. So that kind of thing, though. You can't like say to yourself, oh, this is sandbox. Therefore, I don't have to worry about the web. No. You do have to worry about the web. Like if you're having an agent automatically go out to the web and fetch information for you, you could expose yourself to prompt injection. Yeah. Yes. So they do hide, though, things like tokens. There's a system in the app to obfuscate tokens. There's a system. There's a really nice system where if you want to save a password or an environment variable, there is a built-in environment page where you can secure those variables so that they're never exposed as plain text in the transcript of a conversation. The agent will fetch. It's almost like having the 1Password CLI, which is what I use with codecs on my Mac. The agent never really sees the password in plain text in the chat. It's just using a tool call to authenticate using the 1Password CLI. And what else? So obviously there's browser use powered by WebKit and specifically WK WebView. So the iOS integrations are truly, truly awesome. I mean, the fact that you can search and analyze your photo library. Have it build for you a dashboard based on your health kit data, right? And you can do the classic things that you would do in codecs on the Mac, for example. You can say, hey, I want you to build me a visualization of my heart rate and steps over the past month. And I want you to create it as a web app here inside Minis. And you need to test it. And so you will see the agent in a thumbnail preview at the bottom left of the screen. You will see the agent obviously gather the data via the terminal and then start assembling a web app in the browser and debug the design and the interaction of the web app. You can expand that little preview card to see what the agent is doing. And at the end, it'll tell you, here's your web app. And you can open that web app, which runs on local host in a local web server inside Minis, which can also spawn up a bunch of local servers and HTTPS and Python and whatnot. And then you can choose, oh, I actually want to open this web app and I want to save it to my home screen, for example. You can also do that. But you can, obviously, the other really powerful component I think of Open Minis is that you can organize your different models and different providers in groups and groups of agents and sub-agents. So, for example, I have my default group where I added a bunch of models from different providers that I typically use as my main models, like GPT-5-6-SOL, 5-6-SOL in fast mode via my API key, Kimi K3, and GLM 5.2 provided by Together AI. That's my default group. And at any point, you can switch between your default providers. Is that the purpose of having a group is that it's fast switching? Is that the idea? Fast switching. And you can also enable a, like, you can choose between different modes of allocating tokens to different models. So, the default is a fallback system where if you exhaust your tokens with your first default model, you can automatically fall back to the second one in the list and so forth. Right. Makes sense. Or you can, like, allocate requests. It's almost like a built-in router system where you can allocate requests to multiple models at the same time. I do not use that, but it's possible. So, you could have, you could ask, you could have a prompt that's like, I want to work on this coding project that I'm doing. Use Kimi K2 for, or 3, for the coding part, for this part of the coding project. Use Codex Sol for this other part of the project. You could do that, basically? Yeah. Will be exposed to the other powerful part about Open Minis is that Open Minis knows about itself and knows about its own configuration. So, an agent running in Open Minis can tweak Open Minis itself. That's a little different than a lot of tools out there. It's disabled by default, I think, this behavior. Yes. And you need to enable it in settings. Yeah. But you can ask a model, for example, once you enable this option, you can ask a model to, hey, can you add this new provider to Open Minis? And it'll add a new model provider to Open Minis. Or what the other thing that it can do is there's a use model action where you can ask Open Minis, hey, I want you to work on this GPT Sol. You're going to be the orchestrator, but I want you to use some, I don't know, Claude Fable subagents for the design. And the agent will orchestrate those subagents in Open Minis. Again, those are things that you're probably used to hearing from, like in Codex. On a Mac. On a Mac or Claude Code, this stuff can be done in Open Minis on an iPhone. Yeah. And that, I think, to me, is like the thing that really made me go, oh, I see what's going on here, is that most of these third-party agent chatbot type apps, they're just chatbots. They're a lot like using chatgpt or Claude without any of the tool calling. A lot of them, I mean, some of them have some level of tool calling, but this is really like using Claude Code or Codex right inside a Mac. You've got all that stuff available to you, really. Yes. Yeah. And the special stuff that's iOS only. That's what's interesting to me about it is that, like, a lot of iOS features were kind of locked away from using with agents until something like Open Minis came along. Now you can access those sets of data and those apps in the same way that you would do on your Mac. Yeah. Yeah. Obviously, it goes without saying you can connect any MCP server you want. Yeah. You can install any CLI you want, right? Skills, all that stuff. What you cannot do, I'm trying to think. Like, for example, Open Minis doesn't have a plug-in system. Like, you cannot install, like, plug-ins for Codex or plug-ins for Claude Code in Open Minis. You can install CLIs. You can create skills. You can update those. Again, something else. You can use APIs for things like Claude Code or Codex. You can ask Open Minis to update one of its own skills, right? And all of this is, again, living in the files app. So you can tweak and inspect and work in that workspace locally on your device. Open Minis can sync with iCloud. You can choose what to sync with iCloud. You can set up a limit for file sizes in the iCloud sync. You can see your multiple connected clients. The developer is working on better voice integration. That's one of the things that I'm missing from Open Minis right now. They're working on having multiple voice providers for dictation, as well as having Open Minis read its own answers to you with a voice, like with an 11-labs voice, for example. And you would be able to put in, like, your own API key for 11-labs and use your own voices. Yes, which I did in the beta version. But what's missing is I would like to see a proper voice mode in Open Minis. There are, I believe, OpenAI even has a GPT real-time provider. And so I would love to see a proper voice interface for Open Minis. Yeah, that'd be interesting. But right now it's primarily text-driven. Do you think there would be any advantage to also having a Mac app that would interface with the iOS app? Or is enough of it able to be pulled through via APIs and stuff that it doesn't make much difference? That's a good question. I was thinking about this the other day, and I was kind of stumped at coming up with anything that I would want from my Mac. I guess, I don't know. It's an interesting question whether they're where the roadblocks are with the Mac, perhaps. Yeah. This is true. I would just say that... This episode of App Stories is brought to you by Scribe. A lot of teams don't struggle to scale because they're short of people or budget. They struggle because important knowledge isn't documented, and documenting it manually is a job that rarely gets done. The result is inconsistent work, tools that never get fully adopted, and valuable knowledge disappearing when people leave. And that's where today's sponsor, Scribe, comes in. Scribe is a workflow AI platform that's trusted by 94% of the Fortune 500. It watches you complete a workflow, captures it as you go, and automatically turns it into clear documentation. That means no writing everything up by hand, no stitching together screenshots, and no recreating the same onboarding docs every time someone new joins the team. This kind of thing happens to me a lot. I'm working on a workflow for some new project, and I think I ought to document this for other people to do the same thing. But I'm busy, and I don't have the time, and it falls through the cracks. This is exactly the kind of process Scribe can capture as you work. It's easy because you just turn on the extension or desktop app and do the process as you normally would. Scribe builds the guide as you go, capturing every click, step, and screenshot automatically. What would have taken hours of writing, recording, and cleanup can be done in under a minute and be ready to share. Scribe automatically redact sensitive information, names, account numbers, emails, from every screenshot. As an admin, you can enforce this across your entire team, feel confident nothing slips through the cracks. Anyone following this process can launch real-time, on-screen guidance that shows exactly where to click, step-by-step, inside the actual tool. It's there the moment that Scribe is created and ensures everyone does critical processes correctly. Scribe will suggest improvements to the underlying workflow, not just document it. Once you can see how a process is actually being done, you can identify where steps are redundant, along with where people are getting stuck and where things could be automated or simplified. To see what Scribe could look like for your org, head to scribe.how.appstories and mention us for your first month of Scribe Capture free on select plans. That's S-C-R-I-B-E dot how slash appstories. Our thanks to Scribe for their support of the show. What's truly remarkable here is seeing all these things that we take for granted on a Mac just being done on a phone. I agree. The ability to have a real agent that is sandboxed, sure. It cannot invoke apps outside of its own sandbox. It can open URL schemes. You can't do computer use, for instance, like the equivalent of computer use. Exactly, exactly. Cannot do the equivalent of a computer use because there's no device use framework on iOS. But something that I've started doing, for example, I have Open Minis regularly check. I treat my Mac Studio as the source of truth for all of these things. Like my single point of truth is my codex set up on the Mac Studio. And from there, I use SyncThing, for example, to sync my agents.md file or my codex skills to my MacBook Pro, to my Linux machine, which we'll talk about in App Stories Plus. Yeah. But now with Open Minis, I can also do things like, hey, I want you to SSH into the Mac Studio. And I want you to make sure that you're using the latest version of a skill, right? So keep yourself in sync with the Mac Studio so that the skills that we have here inside Open Minis are also matching the skills that we have on the Mac Studio. And I do find it that by having this essentially shared setup between codex and Open Minis, for most research tasks or things that involve iOS integrations, I just reach out to Open Minis now instead of opening Codex remote. I still use Codex remote for like more complicated things. Sure. And things that involve a desktop browser. Or like ongoing projects, I assume. Ongoing projects, coding stuff, like that all flows through Codex remote. Yeah. But for like web research, iOS stuff, I'm just using Open Minis and it's great. Yeah, it's very good for that. I have two other things I want to cover with you before we do the pro show. The first one is shortcuts. Give me a sense for what you can do with this app with shortcuts and how that kind of empowers the whole thing. And I also think it's worth mentioning that Open Minis has, I think they have an Android app already or they're working on it. Or it's in beta. I think maybe they do computer use on Android. They do some things on Android that you can't do on iOS. With shortcuts, you can just pass a prompt and have it run in the background for up to two minutes, I want to say, maybe. So I just have my automation that refreshes some. I have an MCP server that's running on the Mac and refreshes the authentication token every few hours. So I have a shortcuts automation that gives a prompt to Open Minis on the iPhone to say, hey, talk to the Mac, refresh that MCP token and just then exit. That's the sort of thing that you can do. I have a shortcuts automation that tells Open Minis to... It's something that I stole from the Hermes agent, the self-improvement loop. I basically have this prompt that runs via a shortcut and tells Open Minis, hey, I want you to look into your recent sessions with me, see if there's any new knowledge, any new skills that you want to save for the future so that you become smarter and more personalized over time. I tried setting something like that up and I had trouble with getting it to run when my iPad was closed and locked. Interesting. I don't know what was going on. I got to revisit it because I don't think I set it up the way I wanted to anyway. But it is kind of interesting that you can possibly do that kind of thing, especially when you're doing it remotely and you're not having to be at your iPad or iPhone to do it. The other thing I wanted to talk about a little bit as we close out Federico is Apple and App Intents and kind of how these things relate to each other and whether Apple would be better off kind of going the direction of something like this developer. Because what we're talking about is a system that one person has created. It's been a lot of work, obviously, but one person accomplished this. And it does significantly broader things than App Intents can today in terms of all the tool calling and all that stuff. And so I do wonder whether something more like an existing technology, like creating CLIs from your APIs, is a better route for automating this stuff than inventing a whole new thing called App Intents and doing it that way. No, I think when they will inevitably make a Siri agent, I think they will need to do what this developer has done on their own, but at a system-wide level. I think Apple will realistically come up with their own thing that runs in the background and effectively matches App Intents to a CLI that will not be exposed to users at all. Yeah, not the way this terminal is. No, no, no, no. If anything, because these models, I mean, it's not like Apple is reinventing the wheel here. They're literally using a model distilled from Gemini. These models, they understand command line interfaces so well. MCP is too verbose to, I mean, it's getting better, but it's nowhere near as intuitive for a frontier model as a command line interface is. Command line tools tend to be faster too, I find. Tend to be faster, tend to be less verbose, tend to be more easily debuggable. More efficient, more token efficient, faster. So they will inevitably end up using some kind of CLI bridge for App Intents. They will have a translation layer that matches the entities and the schemas from the App Intent and the enumerations, all those things to a CLI. If anything, because right now, the thing that they, in theory, they have, I haven't been able to actually get it to work, but the way that Siri AI works with third-party App Intents now is the least deterministic thing ever. They're basically having their system prompt, and I'm not making this up. You can just go on your Mac, find the Siri AI system prompt, and read it for yourself. They're literally saying the model, hey, we're giving you some JSON, and this JSON contains some App Intents, and we want you, please, to return more JSON. Like, that's not great. And I fully expect to see plenty of issues and plenty, I mean, happy to be proven wrong? I don't think I will, but happy to. No, especially once the third-party developers get on the bandwagon, because you're going to have so many, you know, like, so there are going to be people with five notes apps and three music apps, and it's just going to be a mess. I also think it's going to be a mess. I think it's a little miracle that they're able to get it to work now, and even then, we're getting some hallucinations now. There's literally a part in the system prompt for Siri AI that says, if you are unable to achieve a goal requested by the user, do not invent a solution. And a few days ago, I just had Siri AI say, hey, if you want, I can copy things to the clipboard for you, and I said, yes. And then I thought about it and was like, no, I do not have the DVD to copy to the clipboard. So, like, these models will, especially an Apple Foundation model distilled from Google Gemini will absolutely ignore its instructions very often. So, we'll see, but I'm very skeptical of third-party app intents working at scale, but hey. I think it's useful to continue explaining the kinds of things that work and don't work with the Siri app. I think most people in our niche think that Siri AI is a magical, like, thing that only Apple can do. And I'm here to tell you that it's based on so much duct tape all over the place and a giant system prompt that is prompting a model like you would prompt a model two years ago. It's kind of primitive. It's primitive in that sense. It's very primitive. And most people think, oh, look what Apple is doing. This is perfect. It really is not. It's a miracle that it's working the way that it's working. But you look at the system prompt and realize, oh, this is how you used to prompt Cloud 4 18 months ago. Yeah, yeah, for sure. Now, you know, the little miracle I had the other day was I was pulling out of my garage and I remembered that I had a text message from my dentist with an appointment details in it. And I had forgot to put it in my calendar and I was driving. I couldn't do that with my phone. So I actually used CarPlay and got Siri to take that text message with that appointment and put it in my calendar and it worked. No, no, it's nice for things like that. Which is a very simple thing in the grand scheme of things. But, you know, the fact that it could do that was useful to me because I was driving and it actually then displayed the appointment on the screen after it did it. So I was able to tell it happened. Yet at the same time, I took an email I got from somebody in German the other day, put that in the Mac app, and it told me that I could open some app that didn't exist on my Mac to do the translation. It didn't do the translation itself, which I was surprised at given that most models are, all models can do translation essentially. Anyway. Anyway. Well, that's Open Minis, folks. It's kind of a wild app. It's a lot of fun to play around with and experiment with. Be careful out there, of course, as always with models in the web and all that stuff. But I don't think that on balance that this is any more risky than any other agent that you might be messing around with on your Mac or elsewhere. So if that's your thing, go for it. It's pretty nice. And with that, Federico, I think we have another episode under our belts. I want to thank our sponsors for this episode. That was InnoReader, Claude, and Scribe. And you can find me and Federico over at MacStories.net and on social media where Federico is at Viticci. That's V-I-T-I-C-C-I. And I'm at John Voorhees. J-O-H-N-V-O-O-R-H-W-E-S. Talk to you next week, Federico. Ciao, John.