From a spare Windows box to a self-hosted AI household.
Monty didn't arrive fully formed. It grew, week by week, from a simple experiment into a genuinely capable assistant woven through everyday life. Here's how it happened.
The arc
What started as "could we run our own assistant?" became a question of how far we could take it, and the answer turned out to be: all the way into the walls of the house.
It began with a simple idea: a private assistant that actually knew our context and answered only to us. It got a name, a personality, and a remit, manage personal life and the Tetherra side of things, keep work separate.
The early prototype ran on a borrowed Windows machine. It quickly earned a permanent home on a dedicated Apple Mac Mini, quiet, efficient, always on. The whole assistant migrated across, and the old box was retired.
Monty learned to talk over secure chat, to remember things permanently across conversations, and to keep an eye on the calendar and inbox. Voice notes went in; spoken replies came back out.
As Monty took on more, security became the priority. Everything moved behind a self-hosted, zero-trust mesh, no open public doors, least-privilege access throughout, and a full security audit to prove it.
The biggest leap: Monty reached into the house itself. Lights, sensors, bins, shopping, all wired into a single brain, controllable by voice or by tap, with a bespoke display on the kitchen wall to tie it together.
Today Monty is simply part of how the household runs, proactive, reliable, and entirely ours. The experiment became infrastructure.
What we learned
Owning the hardware, the data, and the memory changes everything. There's no platform to trust because there's no platform in the loop.
Each capability stacked on the last. Get the foundation solid, then reach further, messaging, memory, security, then the home.
The more an assistant can do, the more it must be locked down. Zero-trust wasn't an afterthought, it was the enabler.