I Built a Local AI Assistant Instead of Cleaning My Office


It started with a single thought:

“What if I could just talk to my computer and it did stuff for me?”

That one question sent me down a rabbit hole of Whisper.cpp builds, Vue-powered dashboards, system prompts, and custom vector memory… all to avoid cleaning the literal pile of Legos, cables, and half-read Stoicism books on my desk.

What I Built

  • 📋 Real-time transcription via Whisper.cpp
  • 🧠 GPT-powered assistant that answers, reflects, or throws shade depending on my mood
  • 🎙 Voice output using local text-to-speech
  • 📁 Context-aware memory using local vector storage (Chroma, for now)
  • 🧪 Experimental “JarretOS” interface via Electron + Tailwind + Vue

It works. Mostly. Sometimes it gives me deeply emotional summaries of my grocery list. That’s part of the charm.

Why I Built It

  1. I wanted something local, not cloud-dependent.
  2. I wanted to build something for myself, not a startup.
  3. I wanted a reason to feel productive without shipping a SaaS or doing marketing.

Also: because a real assistant would cost money. Mine just costs sanity.

The Stack (for fellow nerds)

  • Vue 3 + Tailwind + Electron
  • Whisper.cpp (FFmpeg hell was involved)
  • OpenAI GPT-4 via API or Ollama locally
  • ChromaDB for embeddings
  • A ridiculous bash script that restarts stuff when it dies

Screenshots? Soon. Maybe.

Right now, it works. That’s all I wanted.


You don’t have to build something for the world.
You can build something for your desk.

– The Lazy Engineer 🛠️