How AI was used to build ZenVizor.

ZenVizor was built with Claude Code, under a process I kept firmly in human hands. I set the architecture up front, broke the work into milestones with clear acceptance criteria for each, tested every milestone on a real Windows machine before moving on, and reviewed every change. The repository keeps the receipts: a product spec, a phased build plan with both automated and manual checks, a written set of rules the project cannot break, and the verification records preserved at the v1.0.0 tag.

The app you install has no AI in it. There is no model on disk and nothing performing inference while it runs, it makes no call to any cloud service or remote API, and its list of bundled components includes no AI or machine-learning code. That is not an accident; it follows directly from ZenVizor's first rule, that it sends no traffic of its own. An AI feature talking to a remote service would break that rule outright, and a local model would still show up in the component list. Neither is there.

I think this is the honest way to use these tools: as something that helps a person build carefully, not as a black box bolted into the product. Both halves of that, the careful build and the AI-free result, are open for anyone to verify in the source.

Browse the source on GitHub