Today marks a moment we have been working toward for a long time. We are releasing the very first alpha of Stelix, a private AI assistant that runs entirely on your device.
No cloud. No account required. No data leaving your machine. Just you and your AI.
Why We Built Stelix
Every major AI assistant on the market follows the same pattern. You type a message, it gets sent to a server somewhere, processed by a model you have no control over, and a response comes back. Along the way, your data passes through infrastructure owned by someone else, governed by terms you probably did not read, stored in ways you cannot verify.
We kept asking ourselves: does it have to be this way? The answer, it turns out, is no.
Modern hardware is powerful enough to run capable language models right on your laptop. Apple Silicon in particular has made it possible to do serious AI inference on consumer devices. The performance is there. The memory is there. The only thing missing was an app that made it easy and accessible.
That is what Stelix is. We took the complexity out of running local AI and wrapped it in an experience that feels familiar to anyone who has used a modern chat assistant. No terminal commands, no programming, no config files. Download the app, download the recommended model, and start chatting.
We also noticed that the existing local AI apps on the market are almost exclusively built for technical users. They assume you know what a quantization format is, which model architecture to choose, how much VRAM you need, and how to configure inference parameters. That is a huge barrier for most people. We wanted to build something that anyone could use, regardless of their technical background. If you can install an app and tap a button, you can use Stelix.
What the Alpha Includes
This first release is focused on getting the core experience right. Here is what you will find when you open Stelix for the first time:
Smart Model Recommendations. When you first launch Stelix, it checks your hardware and recommends the best model for your machine. You do not need to research models or worry about whether they will fit in memory. One tap and the download starts.
Offline by Default. Once your model is downloaded, Stelix works without any internet connection. Take it on a plane, use it in a coffee shop with no Wi-Fi, or simply disconnect because you want to. The entire inference pipeline runs locally using your device.
A Clean Chat Interface. We spent a lot of time getting the chat experience right. The interface is minimal and distraction-free. Messages stream in smoothly, and the conversation feels responsive even on longer outputs. We wanted it to feel native, not like a web app squeezed into a window.
Local Storage Only. Your conversations are stored on your device. Nothing is synced, uploaded, or backed up to any server. If you delete a conversation, it is gone. We have no way to recover it because we never had it in the first place.
Privacy Is Not a Feature, It Is the Architecture
We want to be clear about what we mean when we say Stelix is private. This is not a toggle you flip on or a setting you enable. Privacy is how the application is built.
Stelix does not have a backend. There is no server waiting for your messages. Your conversations are never sent anywhere. The app is fully private by default, and any analytics are completely opt-in and anonymized. The only network request Stelix makes is to download the AI model, and after that, you could disconnect your internet entirely and the app would work exactly the same.
This matters because privacy is not just about trusting a company to be responsible with your data. It is about removing the need for trust in the first place. When your data never leaves your device, there is nothing to leak, nothing to request, and nothing to sell.
Built for macOS on Apple Silicon
This alpha is available exclusively for macOS on Apple Silicon. We chose to start here because Apple's M-series chips offer an excellent balance of performance and memory bandwidth for running language models. The Metal GPU framework lets us push inference directly to the GPU, which means faster responses and lower power consumption compared to CPU-only approaches.
We know this limits who can try Stelix right now, and we are already working on expanding platform support. But we believe it is better to ship one great experience than five mediocre ones. The macOS version sets the bar for what Stelix should feel like everywhere.
This Is Just the Beginning
We are calling this an alpha for a reason. It is early. There are rough edges. Some things will break. But the foundation is solid, and we are building in the open because we want your feedback to shape what comes next.
If you have thoughts, ideas, or run into something unexpected, we would love to hear from you. Every piece of feedback helps us understand what matters most and where to focus our energy.
Thank you for being here this early. We are excited about what Stelix can become, and we are glad to have you along for the ride.