Federico Viticci, writing exclusively at MacStories Wednesday:

First, let me share some of the details behind today’s announcement. Sky is currently in closed alpha, and the developers have rolled out a teaser website for it. There’s a promo video you can watch, and you can sign up for a waitlist as well. Sky is currently set to launch later this year. I’ve been able to test a very early development build of the app along with my colleague John Voorhees, and even though I ran into a few bugs, the team at Software Applications Incorporated fixed them quickly with multiple updates over the past two weeks. Regardless of my early issues, Sky shows incredible potential for a new class of assistive AI and approachable automation on the Mac. It’s the perfect example of the kind of “hybrid automation” I’ve been writing about so much lately.

Sky is an AI-powered assistant that can perform actions and answer questions for any window and any app open on your Mac. On the surface, it may look like any other launcher or LLM with a desktop app: you press a hotkey, and a tiny floating UI comes up.

You can ask Sky typical LLM questions, and the app will use GPT 4.1 or Claude to respond with natural language. That’s nice and already better than Siri when it comes to general questions, but that’s not the main point of the app.

What sets Sky apart from anything I’ve tried or seen on macOS to date is that it uses LLMs to understand which windows are open on your Mac, what’s inside them, and what actions you can perform based on those apps’ contents. It’s a lofty goal and, at a high level, it’s predicated upon two core concepts. First, Sky comes with a collection of built-in “tools”1 for Calendar, Messages, Notes, web browsing, Finder, email, and screenshots, which allow anyone to get started and ask questions that perform actions with those apps. If you want to turn a webpage shown in Safari into an event in your calendar, or perhaps a document in Apple Notes, you can just ask in natural language out of the box.

At the same time, Sky allows power users to make their own tools that combine custom LLM prompts with actions powered by Shortcuts, shell scripts, AppleScript, custom instructions, and, down the road, even MCP. All of these custom tools become native features of Sky that can be invoked and mixed with natural language.

Sky is perhaps one of the most impressive Mac app demonstrations I’ve seen since the ChatGPT-inspired artificial intelligence revolution, and it’s what Apple should’ve previewed last year at the Worldwide Developers Conference. Sky is made by Software Applications Incorporated — which has a gorgeous website worth browsing — the team behind Workflow, the app that would go on to become Shortcuts after Apple bought it. It’s no wonder the app is so focused on automation and using macOS’ native tools, such as sending a text or getting information about foreground apps. It’s powered by modern large language models, but they’re not necessarily in chatbot form as much as they are an assistant working on someone’s desktop alongside them.

One of ChatGPT’s most restrictive limits is that any information must be manually added to its context. If someone wanted to ask ChatGPT to summarize a webpage, they’d have to paste the web link into ChatGPT for it to be able to access it. ChatGPT is a chatbot, not an assistant, and we can only add context via text or attachments. Recently, OpenAI has added a “Work With Apps” feature to ChatGPT, using Apple’s accessibility features to look at certain text editors and other apps without having to manually paste text, but ChatGPT can only work with one app at a time, and each one must be enabled separately. It’s hardly ideal, like trying to fit a square peg in a round hole.

Sky uses these same accessibility tools to look into apps on its own. It can even organize files, navigate to webpages, or summarize content because these are all actions exposed by the system, either through Shortcuts or AppleScript. The LLM — Viticci mentions GPT-4.1 — is only the brains behind Sky. It can think and learn how to deal with what it’s been given, but giving it proper context and tools to accomplish common tasks is more of an uphill battle. This was exactly what Apple Intelligence aimed (aims?) to do, but Apple presumably started on iOS, where third-party apps don’t have the same level of access to system functions as macOS. App Intents were the solution on iOS, where developers manually expose actions to Apple Intelligence, but on macOS, Sky can just use the operating system’s existing tools to work in apps.

Sky is the first personal assistant I’ve actually wanted to use on my Mac. A rather public fact about me is that I despise voice assistants. They’re handy sometimes, but mostly, I prefer typing because I type faster than I speak. This is why I love ChatGPT — advanced voice mode is great sometimes, but I can always type to it and receive a speedy text answer I can always reference. (I think human communication stresses me.) If I have a question for Sky about something I’m working on, I can just invoke it quickly and discreetly, much like Alfred or Spotlight, and have it know everything I know. In other words, I don’t have to tell it anything, like what’s on my screen or my current project. It’s like Cursor but for everything on the Mac.

I’m sure Sky will be costly, but it’s the first implementation of LLMs that truly goes beyond a chatbot. A friendly assistant is only one part of the mission to create intelligent computers. I’m on the record as saying that true artificial general intelligence should be able to live alongside us in our world — designed for our eyes, hands, and brains — and any computer that requires our attention to translate the world into something more machine-friendly is miles away from AGI.

Computer code is a rudimentary form of translating human ideas into something a computer can understand. As computers have gotten more powerful (“intelligent”), programming languages have gotten simpler, from Assembly to C to Python. Compilers now understand more nuance without programmers telling them what to do. Example: Swift is a type-inferred language only because the computers that compile Swift are so complex and powerful that they can implicitly infer what type an expression is. But the Swift compiler isn’t an AI system (obviously), and neither is ChatGPT, because both systems still require us humans to tell them about our world.

The Swift compiler knows more about binary than I ever will, and ChatGPT knows more facts than I do, but it’s still up to me to write a program in Swift or tell ChatGPT what I’m working on and what I need its help with. ChatGPT doesn’t know I’m writing something unless I open a new chat and say, “Hey, help me with this text I’m writing.” Swift’s compiler doesn’t do anything unless I give it a valid Swift program I wrote. ChatGPT is the Swift compiler of AI systems — smarter than human work (Assembly), but still requiring manual intervention.

Sky is one step closer to a world where I don’t have to tell ChatGPT that I’m writing something. Like how Swift is type-inferred, Sky is task-inferred, if you catch my drift. I don’t have to translate what I’m doing into something Sky understands. It already has the context required to do what I need it to. That makes it come a step closer to AGI — a system that can work alongside humans without any manual intervention or translation. Sky isn’t renaming files for me, of course, but I don’t have to give it the file names, tell it I want them renamed, and then paste in the new file names it gives me, like I would have to with ChatGPT. It just renames them after I ask it to. There’s still friction — in my Swift analogy, I still have to write the program — but it’s inferring the task I’m trying to complete.

I have no idea if my half-baked programming language analogy makes any sense, but I think it’s a good way to think about these systems as they get closer to market, whether it be Apple Intelligence, Google’s Project Astra, ChatGPT, or Sky.