Custom Applications

The 1001 tools on the rest of this site all share one constraint: they run inside a browser tab, finish in a moment, and never send your data anywhere. That constraint is what makes them trustworthy, and it is also what stops them doing certain jobs at all. You cannot transcribe an hour of footage in a tab. You cannot re-order a page that belongs to somebody else's website.

These are the applications built for the jobs that did not fit. They are separate programs rather than pages, so this section describes them and says plainly what state each one is in — including where that state is "not finished".

Describe & Cut

A Windows application that points AI vision models at a folder of photos and video and writes back something useful: a description of every photo, a shot-by-shot chart of a video's timeline, or a finished edit assembled from a one-line prompt. It runs the models on your own machine through Ollama, so the free path needs no account and no API key.

How Describe & Cut works →

Amazon Rating & Price Sort

A Chrome and Edge extension that re-sorts an Amazon search page in place, by rating band first and price second, while demoting listings with too few reviews to judge. It makes no network requests at all — everything it needs is already in the page you loaded.

How the Amazon sorter works →

Why these are not on the site

Both of them could be described as "a tool", and neither could be one here. Describe & Cut needs ffmpeg, several gigabytes of model weights and hours of processor time; a browser tab is the wrong container for that even before considering that your footage would have to be uploaded somewhere. The Amazon sorter has to read and rearrange a page on amazon.com, which by design no website can do to another. Only an extension, running with your permission in your own browser, is allowed to.

What they do share with everything else here is where the work happens. Neither one sends your files, your searches or your shopping to a server. That is the same reason as ever: data that is never collected cannot leak, and a claim about handling is weaker than a design that makes handling unnecessary.