I want to let everyone know in advance i did use Claude to help me build the code and this description.
Ever since I watched Iron Man as a kid I wanted my own Jarvis, a voice assistant that actually controls your PC, not just searches the web. Everything out there (Cortana, Alexa, Google Assistant) felt like a watered down version of what I had in my head.
What it does
You say "Nexus" followed by a command and it executes it instantly:
• "Nexus, open Geometry Dash" → game launches
• "Nexus, set volume to 40" → done
• "Nexus, what's on my screen?" → AI looks at your screen and tells you
• "Nexus, order a pepperoni pizza from Dominos" → opens the ordering page ready to go
• "Nexus, thermals" → reports your CPU load and GPU temp out loud
• "Nexus, game mode" → closes Chrome, launches Discord, sets volume to 60 in one command
It even handles speech recognition errors — saying "rinecraft" still opens Minecraft.
How it works
The whole thing runs locally on your GPU. No subscription, no cloud, no API key needed.
— Python for everything
— Ollama running gemma3:4b on GPU as the AI brain
— Google Speech Recognition to hear commands
— pyttsx3 for the voice responses
— A custom executor that dispatches OS-level actions based on what the AI returns
When you say a command, it gets sent to the local Ollama model which returns a structured JSON action plan. The executor reads that JSON and actually does the thing — launches the app, changes volume, opens a URL, whatever. The AI never touches your system directly, it just decides what to do and the Python code does it.
It also scans all your drives on first boot and indexes every installed app automatically, so you never have to manually add paths for most things.
Features
✓ Runs 100% offline on your own GPU
✓ Auto-scans all drives and indexes every installed app
✓ Fuzzy speech matching (handles mic errors and accents)
✓ Screen vision — can see and describe what's on your monitor
✓ Persistent memory — remembers past conversations across sessions
✓ Web dashboard at localhost:8080
✓ Food ordering — opens Dominos, KFC, Subway etc. directly to the order page
✓ Macro system — chain multiple actions into one voice command
✓ Adapts to failed launches — learns working paths over time