The suite started in the browser. FastAPI backends, vanilla JS frontends, Ollama on the same local rig. That model still works. But not every job fits a tab.
Email is the obvious exception. You do not want to log into a web app, unlock a database, and wait for a sweep before you triage your inbox. You want an app that sits in the Dock, runs when you click it, and keeps your credentials in the same encrypted database the operating system already guards. So I built one.
inbox0 by MyOwnLifeOS is now pending App Store review. It is the first native member of the suite: a free, local AI email assistant for Gmail, built for macOS.
What Changed
The web version of inbox0 was FastAPI, SQLite, Ollama, and vanilla JS. For the desktop app I kept the same three-agent pipeline but removed the Python sidecar entirely. The Rust backend uses the llama-cpp-2 crate, stores data in SQLCipher, and runs the local model on Apple Metal by default. A CPU fallback is there if you need it, but on a modern Mac the inference stays on your GPU.
The app downloads a 2 GB Llama 3.2 model on first setup, connects to Gmail over IMAP, and triages or drafts replies without ever touching a cloud LLM. Credentials live in SQLCipher. Mail never leaves the device. Same zero-cloud rule as the web suite, now in a native binary.
Beyond triage, the desktop app includes MailMind: the local email knowledge layer from the web version, ported to run entirely on your Mac. Create collections from your mail, ask questions against them, and the answers come from your own inbox on your own hardware.
Why macOS
Apple Silicon is the most consumer-ready private AI platform I have used. You get a fast, efficient Neural/Metal stack inside a machine you already trust. No dedicated GPU required. No Linux driver hunt. No Windows WSL shim. You install the app, it runs, your data stays on device.
This matters for the MyOwnLifeOS philosophy. Sovereignty is not just about open models or local servers. It is about computation happening on hardware you control, with data that never crosses an API boundary to a cloud AI vendor. A Mac with Apple Silicon makes that practical for normal users in a way that a multi-GPU tower is not.
Guide-Coded, Again
The app was guide-coded in a VM on my local inference cluster. Same setup as the rest of the suite: the assistant writes many of the lines, but I make every architecture call, review every diff, and run every build. The difference this time was the target. Shipping a native app means codesigning, provisioning profiles, entitlements, sandboxing, and App Store packaging. None of that is the romantic part. It is the part that turns a side project into software a stranger can install.
First submission happened the same day the idea came up. The build went from idea to App Store Connect in a single afternoon. That is not a brag about speed. It is a measurement of how small the real work becomes once the stack is understood and the decomposition is clean.
MyOwnLifeOS is no longer just a web suite. It is a family of sovereign agents that can live wherever your data lives, including natively on your Mac.
What Comes Next
More native apps. The ones that benefit from a Dock icon, offline operation, and OS-level encryption are the obvious candidates. The web versions will stay. The desktop versions will be the same apps, running closer to your data.
If inbox0 clears review, I will share the App Store link. Until then, the suite keeps growing.