The catalogue holds more than 4,400 entries. Describe what you are looking for in plain language — the engine understands meaning, not just words.
How it works
A conventional search matches keywords: it finds “Tulbaghia” if you type “Tulbaghia”, and nothing at all if you type “fragrant bulb for a shaded pot”.
Here, every entry has been turned into a sequence of 1,024 numbers that places its content in a space where related ideas sit close together. Your question goes through the same transformation, and the engine returns the nearest entries — even when they share none of your words. The catalogue, the index and the computation all stay on a dedicated server.
Search by meaning
Try: “fragrant bulb for a shaded pot” · “tall white flower for a sunny rockery” · “which bulb flowers in December”
What this search does
The engine picks up the criteria you express — fragrance, shade, container growing, rockery, cut flowers, flowering period — and sets aside entries that do not meet them.
No artificial intelligence writes anything here: you get the entries themselves, ranked by relevance. It is instant, and nothing can be made up.
Ask a question
Try: “how should I water my bulbs during a heatwave?” · “what to plant now for flowers at Christmas?” · “which fragrant bulb suits a shaded pot?”
What this answer is, and where it comes from
The search above is the first step: the most relevant entries and growing advice are retrieved, then handed to a language model that writes from them — and from nothing else. The model has no knowledge of the catalogue of its own; it cannot invent a plant or a flowering period. The documents it used appear below every answer, with their links.
The engine also recognises what kind of question you asked: a request for a plant yields a recommendation, a practical question yields advice drawn from the articles published on this site.
Why two engines, and which to choose
Remote engine — the writing is handed to a specialised service running on hardware built for it. An answer in three to five seconds, from a very large model. Only the selected entries are sent to it, never the catalogue.
Private engine — the writing happens on our own server: no data leaves it, no third-party service is involved. It is slower, one to three minutes, because the model runs on an ordinary processor without a graphics card. That is a matter of cost, not a limit of the method: the same setup with a graphics processor answers within seconds. The result itself is already fully usable.
Connect the catalogue to your assistant
If you use Claude or ChatGPT, you can query our catalogue straight from your conversation, without visiting this site. Your assistant will consult the entries and answer in its own words, citing its sources.
https://rag.pbq-system.com/mcp/eb0d80871bbf7802dd2e46d0d207ebfaHow to add it
In Claude — Settings, then Connectors, then “Add custom connector”. Paste the address above and confirm. The feature requires a paid plan.
Then ask something like “which fragrant bulb suits a shaded pot?”: your assistant will search the catalogue before answering.
This is an MCP connector, an open standard supported by several assistants. It grants access to search only: no personal data is sent, and nothing can be modified.