Hey Everybody!

Over the last few weeks I built Haplo AI, an app that lets you download and run large-language models entirely on-device—no cloud, no trackers, just private, offline AI chat wherever you go. It’s perfect for spots with spotty (or zero) Wi-Fi (or if you just don't like handing billion dollar companies all of your data😅).

How it’s built

1. Swift + SwiftUI Front-End
2. Kuzco — my newly open-sourced Swift wrapper for llama.cpp
3. Metal acceleration for fast, battery-friendly inference.
4. Background model downloads that resume even if the app quits.
5. Encrypted SwiftData persistence with live Combine updates.
6. On-device safety filter—no external calls required.

Why I built it

I wanted ChatGPT-level convenience without surrendering my data, and something that keeps working when the coffee-shop Wi-Fi dies. Haplo AI lets me brainstorm, journal, or outline articles anywhere (used 30,000ft in the air with no wifi and it works like a charm☺️)

Cool extras that I'm working on

• System-action agent: booking Ubers, sending texts, auto-scheduling, firing off social posts, etc.
• Open-sourcing more of the “agent” layer once APIs stabilize.

Wins so far

• Reached #64 apps in Apple's Utilities chart (only for a day.. but still 🎉)
• Received more than 100 downloads across macOS and iOS devices

I’d love your feedback on architecture, performance hacks, or UX polish!

P.S. I'm loving Screen Studio for recording my apps to showcase!