Local "Jarvis" AI Assistant

01 · The problem

Cloud AI assistants leak data and rack up API costs. The goal was a responsive assistant that runs entirely on personal hardware with total privacy.

02 · Agent architecture

1. 1

   Integrated open-source LLMs (Llama, Gemma) into a Python runtime.

2. 2

   Put every model call behind one switch, so a local or a hosted model can answer.

3. 3

   Wired natural-language commands directly to on-device actions.

03 · Outcome

An assistant that runs natural-language commands on personal hardware. Honest note: the current code calls hosted models through OpenRouter, so the fully offline mode is not in this version.