Inspiration
ATRI is a lightweight desktop companion that combines a transparent always-on-top character with practical AI assistance. I wanted to explore how a local desktop presence can feel useful without becoming a heavy chat application.
What it does
The project provides a small Windows desktop pet with idle animation, drag interaction, a tray menu, and clean exit behavior. The longer-term goal is a hybrid workflow: local interaction for speed and privacy, plus cloud models for tasks that need stronger reasoning.
How I built it
The desktop shell is built with Python and native Windows APIs. The interface stays lightweight and runs without a full browser runtime. I am prototyping a model router that can use both OpenAI and Claude APIs, with local fallbacks when cloud access is unavailable.
Challenges
The main challenge was keeping the desktop app genuinely light: low idle CPU, low memory, reliable always-on-top hit testing, and predictable drag behavior. Packaging and clean shutdown also needed careful handling.
What is next
Next I will add a small task panel, model routing, and a reproducible demo that shows the hybrid local/cloud workflow end to end.