HA Builder & Home Guardian

The idea

HA Builder and Home Guardian split one tempting “AI for the smart home” idea into two jobs with very different trust requirements.

HA Builder is the maker. It focuses on supported Home Assistant configuration changes. Home Guardian is the watcher: it observes useful home-system signals and surfaces things that deserve attention. Keeping those roles separate is intentional.

Home Guardian gets to be nosy; HA Builder has to fill out the paperwork.

What it does

Home Guardian starts from read-only awareness—looking for meaningful conditions without quietly deciding that every unusual sensor reading needs an intervention.

HA Builder follows a gated build workflow. Its capabilities expand in small batches, with backups, validation, exact-change tracking, and owner approval where a change affects live behavior. Early work intentionally started with safer metadata changes before moving toward bounded script behavior.

Why it matters

A smart home is unusually personal infrastructure. A bad automation is not just a failed test; it can turn off the wrong thing, wake someone up, or create the exciting new household ritual of asking why the lights are doing that.

Separating observation from mutation makes the system easier to trust and easier to expand. Each agent has a clear job and a clear boundary.

That separation also lowers the odds of an innocent sensor oddity becoming an unsolicited home-improvement project.

What I learned

The safest path to a capable agent is incremental. Proving read-only observation, then constrained edits, then backup and verification behavior creates a much stronger foundation than starting with broad device control.

I also learned that “can the agent change this?” is only half the question. “Can it prove exactly what changed, recover safely, and know when to ask?” is usually the more important half.