Schematic · Self-hosted AI · Rev 1

Your AI assistant should live in your house. Not someone's cloud.

A step-by-step blueprint for building a private voice assistant that runs entirely on hardware you own — a local model, real Home Assistant control, no subscriptions, and nothing leaving your network.

Why build it yourself

Big-tech assistants are convenient because they send everything you say to a server you don't control. This is the other way to do it.

Local

Runs on your hardware

Every request is processed on your own GPU. Nothing is shipped off to a third party to be logged and mined.

Connected

Actually controls things

Wired into Home Assistant, so it does real work: lights, scenes, sensors, routines — across multiple rooms.

Yours

No monthly fees

Build it once on gear you already own. No per-seat pricing, no surprise rate hikes, no rug-pull when a startup folds.

What's inside the blueprint

A complete build sequence — written by someone who actually runs this at home, not a rewritten docs page.

  1. Hardware & GPU sizingWhat you need, what's overkill, and how to pick a model that fits your card.
  2. Installing a local modelStanding up the inference backend and serving it reliably on your network.
  3. Local speech in and outWiring up speech-to-text and a natural text-to-speech voice — fully on-device.
  4. Home Assistant integrationConnecting the assistant as a conversation agent that can read sensors and trigger actions.
  5. Multi-room & personalitiesRunning distinct assistants per room, each with its own voice and behavior.
  6. Safe remote accessReaching it from your phone without exposing your home network to the internet.

Questions, answered

The things people ask most before building this. Got one that isn't here? Send it over — it'll probably end up on this list.

What hardware do I need?
A computer with a dedicated GPU is the main requirement — that's what runs the AI locally and quickly. The guide shows how to match a model to the card you already have (or are buying), plus a simple mic-and-speaker setup per room. No server rack required; one capable machine gets you started.
Can I run it without a GPU?
Technically yes, but responses get slow enough to be annoying for everyday voice use. The guide is upfront about this and helps you find the most affordable GPU that still feels responsive, rather than overspending.
Which smart-home system does it control?
It's built around Home Assistant, the open, self-hosted hub most privacy-minded setups already run. If your devices work with Home Assistant — and the large majority do — the assistant can control them.
Does anything get sent to the cloud?
No. The model, the speech recognition, and the voice all run on your own hardware. Your conversations and your home data never leave your network.
Do I need to be a programmer?
No, but you should be comfortable following technical steps — editing a config file, running a few commands in a terminal. If you've set up Home Assistant before, you're well within range. Every step is written out in full.
What does it cost to run?
After the hardware, essentially just electricity. No subscription and no per-request fees — which is the main reason a local setup pays for itself over time compared with a cloud assistant.
How long does the build take?
A focused weekend gets you a working single-room assistant. Multi-room setups and custom personalities are things you layer on afterward, at your own pace.

Early access

It's being finished now. Join the list to get it first — and at the lowest price it will ever be.

$29 $49
Launch price for early-access readers · one-time, no subscription