There are two ways to give a site good search. Send your content to a service built for it and read the results back, or keep the whole thing inside your own site. Most of the market picked the first. Fetchie picked the second, on purpose.
What a vector database is, in plain terms
You’ll see AI search and chat tools talk about "vectors" and a "vector database". It is worth a plain explanation, because nobody gives you one and you should not have to already know.
To match on meaning rather than on the exact words someone typed, a tool first turns every page on your site into a long list of numbers, called a vector (or an "embedding"). The numbers are meant to capture the gist of the page, so two pages about the same thing end up with similar numbers. When a search comes in, it is turned into numbers the same way, and the tool looks for the pages whose numbers sit closest. Finding those closest numbers quickly is a specialised job, so the lists are kept in a store built for it: a vector database.
The work is in everything that store needs around it. The numbers have to be generated in the first place, usually by an AI model charged per item. They live in a separate database, often a hosted service with its own monthly bill. Every time you change a page the numbers have to be made again and the store updated, or search keeps answering from the old version. And the service has to be kept running and warm, so the first search after an idle stretch does not arrive cold.
In tiny-house terms, it is like keeping a second, coded copy of everything in your house in a rented storage unit across town. Move something at home and you drive over to update the copy, and you pay rent on the unit whether you visit or not.
There are sites where that is worth it, when matching on meaning is the whole point. For finding the right page on a WordPress site, it is a lot of moving parts for a job that matching the words people actually type, ranked well, already does.
What "no external service" means
Fetchie’s index is a small set of tables in your own WordPress database. When someone searches, the answer is worked out there, on your own server. Nothing is sent out to be searched, and no results are fetched back in.
There is no account to open with a search vendor, no vector database to stand up and feed, and no API key whose quota you have to watch. The search box in your header talks to your own database and nothing else.
No vendor invoice, ever
A hosted search service is a subscription. It is usually metered too, on the number of records you index or the number of searches your visitors run, so the bill grows exactly as your site succeeds.
Fetchie is a flat yearly licence and then nothing. Your traffic can double, your catalogue can triple, your best month can break records, and none of it changes what search costs you. There is no second bill arriving from somewhere else.
Nothing to keep in sync
When the index lives elsewhere, it is a second copy of your content that has to be kept in step with the first. Publish a post, change a price, unpublish a product, and each change has to be pushed out and confirmed. When that sync lags or breaks, search keeps showing the wrong thing until someone notices.
Fetchie has no second copy. The index and the content are the same thing, in the same place, updated together the moment you hit save. There is no sync to fall behind and none to break.
Cheaper to run, and to keep running
The licence is the obvious saving. The quieter one is everything that is not there.
No external account to provision when you build the site. No API budget to reconcile each month. No embedding costs that run every time you edit a page. No integration to re-test when the vendor changes their API, and no outage on their side to explain to your customers. Search that lives entirely on your own server is one fewer moving part, and over the years a site runs, one fewer thing to service is worth as much as the fee you never pay.
Lean by design
Keeping it in-house only works if it stays light, so Fetchie is built to. The index is a handful of small tables, a few megabytes for a typical site, which is why it runs comfortably on ordinary shared hosting rather than needing a search server of its own.
The result is search that behaves like the expensive kind — ranked results, highlighted snippets, deep links to the passage — while staying small enough to sit inside the site it serves.
Your content stays yours
There is a data point to this as well as a cost one. Your content, and the record of what your visitors search for, never leave your hosting. Nothing is copied to a third party to make search work, because search was never sent anywhere in the first place.
That is the whole idea: your site, your search, your server, your bill and no one else’s.