I wanted to share how ZecHub AI is wired into search on the ZecHub Wiki.
Two ways to find answers
Classic wiki search
The search modal still works like a normal wiki lookup: you type a query and get matching pages (titles, topics, paths). You can move through results with the keyboard and open a page as usual. This is fast and fully on-device for the index side of things - no LLM required.
“Answer with AI”
There is a separate tab in the same search UI: Answer with AI. There you ask in natural language. The backend:
Runs semantic search over pre-embedded wiki content (vector similarity over chunks stored in Supabase).
When relevant chunks are found, the model answers in a RAG-style way: answer grounded in those excerpts, with source links back to wiki pages.
When nothing matches well enough, it can fall back to general knowledge about Zcash / ZecHub (with a note that the live wiki is the source of truth).
Off-topic questions get a polite out-of-scope reply instead of a random answer, so the assistant stays focused on Zcash and ZecHub.
So: search helps you navigate to pages and “Answer with AI**”** helps you get a synthesized answer (with sources when the corpus has them).
Tech stack (high level)
OpenAI for chat and embeddings.
Supabase for storing and querying document embeddings used by semantic search.
Next.js API route that ties search → context → model response.
Planned direction:
Right now, semantic search depends on embedded wiki chunks stored in the database. As the wiki grows and pages change, I want a reliable, automated pipeline to refresh those embeddings so answers stay aligned with the latest content.
The feature is open source in the ZecHub Wiki repo and zechubai for anyone who wants to review or reuse ideas.
Would be nice if the “Wiki” were an actual wiki instead of being falsely advertised as such.
I found an error, and no I’m not creating a GitHub or Discord account.
How about just ZecHub? With zechub.org redirecting to zechub.wiki, it seems there’s no difference between ZecHub and ZecHub Wiki. The zechub-wiki repository seems there for the sake of organising content… so why not call it zechub-content? That would not be a misnormer.
Also, the zechub repo description reads:
A decentralized education hub for Zcash.
What’s decentralized about it? As far as I can tell, it’s a single web page on a single server. Having a DAO does not make the site itself decentralized. All in all, I get the sense people are just throwing buzzwords around instead of providing accurate descriptions.
And what’s with all the trackers? This is supposed to be a privacy-centered community. It is thinks like this that effectively make it a honeypot.
I saw that - among the wallets - Exodus isn’t tagged as having Linux support and both Guarda and Atomic aren’t listed at all. Thus Linux + Multi Coin yields 0 instead of 3 results.
Having a Free Software/Libre/Open Source tag would also be nice.
RAG over existing wiki content is a practical approach. Two search paths (keyword and semantic) lower the barrier for people who don’t know exact terminology yet. Worth watching as the corpus grows: how embedding refresh frequency affects answer accuracy for newer protocol changes.
