An AI support assistant has to find the right part of your site before it can answer from it. Most tools do that finding with vectors and a vector database. Helpie does not, and the difference shows up on the bill and in the upkeep.
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 question 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 answers keep coming from the old version. And the service has to be kept running and warm, so the first request 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.
How Helpie finds the answer instead
Helpie keeps a plain index of your words and where they appear, right inside your own WordPress database. When a question comes in, it finds the passages that match and hands them to the AI to word the answer. The finding is done by your own index; the wording is done by the model. No vectors are generated, and there is no vector database anywhere in it.
That means nothing to host, nothing to feed, nothing to keep in sync, and nothing to keep awake. The index updates itself when you publish, and it stays small: a couple of hundred pages is a few megabytes, which is why Helpie runs on ordinary shared hosting rather than needing a service of its own.
Why we built it this way
We had run the other kind: a separate store you host, feed, and keep alive with a scheduled ping so it does not fall asleep between questions. For something a business installs and then leans on every day, that upkeep is a liability rather than a feature, and it is the part most likely to break on the slow afternoon when nobody is watching.
So Helpie answers from an index that lives in the same place as your content and looks after itself. One system, on your own server, with your own content, and nothing to babysit.
What you give up, and what you don’t
Vectors earn their keep when matching on meaning is the whole point of the product. For answering a customer’s question from your own pages, the harder problems are finding the right passage and refusing to guess when there is not one, and Helpie handles both without the extra machinery: it grounds every answer in your content and cites the page it came from.
So you keep the grounded, cited answers. You skip the second database, the re-embedding costs, the sync, and the keep-alive. For a support assistant that a small business runs on its own site, that is the trade worth making.