HouseCommand AI // Estate-grade operations

Command your home like critical infrastructure.

HouseCommand AI gives high-end residences a local-first operations layer for power, water, climate, security, network, and automation — with deterministic safety gates and human-confirmed control.

Install → View System Architecture

Local-firstPermission-gated Human-confirmedVerified control

// every demo below is computed in your browser — solved, not played

SCROLL

// 00 · Live control theatre

Solved, not played.

A two-dimensional heat field — ∂T/∂t = α∇²T + q — computed live in your browser over a model estate. Four zones hold setpoint by deterministic PI control while a winter night leans on the envelope. Nothing below is a recording.

LIVE SOLVE EQ ∂T/∂t = α∇²T + q GRID — 60 FPS 0.00×10⁹ CELL UPDATES OUTDOOR —

DRAG INSIDE A ROOM — YOU JUST OPENED A WINDOW

The controller you are fighting is the point. Deterministic loops hold the house. Disable heating and watch on-prem freeze protection take over at 42°F — no cloud, no model, no permission needed to keep pipes from bursting.

// numerics — ∂T/∂t = α∇²T + q, forward-time centred-space on a 192×120 grid, 3 substeps per frame; diffusion numbers λ = αΔt/Δx²: air .235, glass .09, wall .02 — all under the 2-D stability bound λ ≤ ¼. Zone control: PI, k_p=.09, k_i=.0016, integrator clamped [0,.85]; freeze-protect latches at 42°F, releases at 46°F. Diurnal forcing T_out=39+13·sin(ωt).

// 00b · In plain terms

What it actually does.

Strip away the jargon and HouseCommand is three ideas working together. An AI can suggest actions for your home — but it is never allowed to carry out the ones that matter without a person, and everything it touches is written down.

The AI proposes

You talk to your house in plain language — "turn on the porch light," "is anything wrong outside?" The AI works out what you mean and suggests the action. It only ever suggests.

A safety engine decides

Every suggestion meets a fixed rulebook. Harmless things (lights, temperature) just happen. Consequential things (locks, water, power) need your confirmation. Dangerous things the AI simply cannot do — the button does not exist.

Everything is recorded

Every action — done, refused, or merely suggested — is written to a tamper-evident log. If anyone edited the history, it would show. Nothing happens quietly.

An analogy. Think of the AI as an extraordinarily capable house manager who can prepare anything but is never handed the keys to the front door, the water main, or the breaker panel. For those it must ask you — and a notary in the corner writes down every request, every answer, and every action in permanent ink. Replace the manager with a worse one, or a dishonest one, and the locks, the rules, and the notary do not change.

And underneath all of it: every physical switch, key, and valve still works by hand, and a single “AI hold” pauses the AI entirely while the house keeps protecting itself.

// 03 · The HouseCommand difference

Governed control, end to end.

Eight principles the platform enforces in software — not policy, not documentation. The engine, not the model, decides what executes.

// 04 · Command domains

One layer. Every system.

DRAG-FREE · SCROLL TO TRAVERSE