AI-Buzzby Nick

Nick’s AI-assisted home workshop

Can I automate my life without turning the automation itself into a second job?

I use AI agents to help build, troubleshoot, and maintain local, owner-controlled alternatives. Then I live with them long enough to learn whether they save more work than they create.

See what’s on the bench

On my bench Building

House AI

Can a locally processed voice assistant be good enough to use every day?

What works

The Windows PC can transcribe speech, stream a Qwen answer, generate spoken audio, and keep a short follow-up conversation. Automated tests also cover the exact, source-backed household-answer rules.

What remains open

The spare Galaxy, the complete Galaxy route, a real timing baseline, retention across every component, and ordinary home use are still untested.

Evidence
Bench-tested
Next test
Build and run the repeatable Windows quality baseline for local voice turns.
Open the House AI case study

What changed recently

One useful change, with the missing proof left visible.

Turn instrumentation is in place

The browser and Windows gateway can now join a content-free turn record from recording through playback, with optional local JSONL output. No real baseline or Galaxy timing has been collected. Next, I’ll build and run the repeatable Windows quality test.

Read the build note

What the repository record shows

The task was precise. The agent’s role wasn’t recorded.

A private issue and pull request record the turn-instrumentation task and merged result. Neither identifies whether an agent participated or separates any agent work from Nick’s edits.

Task recorded
Trace one turn from the browser’s recording action through playback or failure.
Merged result
Added browser and Windows-gateway timing stages plus failure and cancellation records.
Nick still has to
Operate the physical phone and Windows PC and judge the voice experience.
See the contribution evidence boundary

The maintenance tax

If it saves five minutes and creates two hours of upkeep, it failed.

Every build inherits updates, broken integrations, credentials, network failures, recovery steps, and the burden of remembering how it works. AI-Buzz treats that ownership cost as part of the result—not a footnote after launch.

House AI has not reached ordinary use, so there is no long-term maintenance history to dress up. When that evidence exists, it will sit beside the demo and the code.

How I judge a project

Valid outcomes

Keep it, simplify it, buy the supported option, go manual, or stop.

A project does not have to survive to be useful. It does have to tell the truth.