Every AI chatbot plugin makes the same promise: answers from your content, not generic internet mush. Very few explain how that actually works, and even fewer let you do it for free. This guide covers both. First, a plain-English explanation of RAG, the technique behind every “trained on your content” claim. Then, a full setup of Hyve Lite, a free WordPress AI chatbot that answers visitor questions using your own posts and pages.
One disclosure up front: Hyve is our very own plugin, built by Themeisle (the company behind WPShout), so expect full honesty about the limits, including what the free version won’t do. Here’s what’s ahead:
What “trained on your content” actually means
Let’s clear up the biggest misconception first: nobody fine-tunes a model on your blog. There is no training run, no custom model, no week-long GPU bill. What these chatbots use instead is RAG, short for retrieval-augmented generation, and the idea is simple enough to explain in three steps:
- Your question becomes a number. When a visitor asks something, the chatbot converts the question into an embedding, a numeric fingerprint of its meaning.
- Your content gets searched by meaning. Your posts and pages were already split into small chunks and converted into the same kind of fingerprints. The chatbot retrieves the chunks closest to the question.
- The AI writes the answer. The best-matching chunks go to the AI model together with the original question, and the model writes a reply grounded in your actual text.
Two practical consequences follow. The good one: the bot answers from what you really published, so update a post, re-sync, and the answers change. The sobering one: it’s only as good as your content. Thin posts in, thin answers out.
What you need before starting
- A WordPress site with real content. Published posts and pages are the raw material. The more genuinely informative they are, the better the bot performs.
- Hyve Lite. The free version on WordPress.org (7,000+ active installs, requires WordPress 6.2+ and PHP 7.4+). It’s open source (GPLv3), with the code public on GitHub.
- An OpenAI API key with paid credit, if you power the bot with your own key. Free-tier keys tend to fail with “Too many requests” errors, so add a few dollars of credit first. Skip this entirely if you go the Hyve Connect route.
- A Hyve Pro license, if you go the Hyve Connect route instead. Pro is a one-time payment: €99 lifetime for one site or €279 for 50 sites, with a 30-day money-back guarantee.
- Optionally, a free Qdrant account. Only relevant if your knowledge base outgrows the free 500-chunk cap. More on that below.
Two ways to power the bot: Connect or your own key
Before clicking anything, decide where the AI actually comes from, because Hyve offers two routes and they suit different situations:
- Hyve Connect (Pro) is hosted AI included with your license. No OpenAI account or API key to create; the plugin talks to a hosted service instead. Limits pool across your licensed sites: 10,000 knowledge base chunks and 10,000 chat messages per month. Pick this for zero key management or multi-site setups.
- Your own OpenAI API key (works with the free plugin). You create the key, pay OpenAI directly for usage, and face no limits from the plugin developer. Pick this if you already have an OpenAI account with credit, or want zero plugin cost.
Everything else in this guide works identically on both routes; the choice only changes what you click in Step 2.
Setting up a WordPress AI chatbot on your own content
The whole setup takes about fifteen minutes, most of it spent picking the posts for the knowledge base.
Step 1: Install Hyve Lite
In your dashboard, head to Plugins > Add New, search for “Hyve”, and install and activate Hyve Lite. Activation adds a new Hyve menu to the sidebar. Everything happens in there; no coding required at any point.
Step 2: Pick your AI – Connect or your own key
Going with your own OpenAI key? Head to Hyve > Settings > Advanced and paste the key into the API Key field, then hit Save. You can create a key in your OpenAI account under API keys. Chat data then goes straight from your site to OpenAI, and every visitor message consumes a little credit.
Going with Hyve Connect (Pro)? Skip the API key entirely. With an active Pro license, open the Hyve Connect panel and connect; the OpenAI key settings simply stay inactive. The panel shows live usage meters against your pooled limits.
Step 3: Build the knowledge base
Open Hyve > Knowledge Base and click through to Add Data. You’ll see a list of your posts and pages with an Add button next to each title. Pick the content that should power the answers.
Resist the urge to add everything. This selection shapes how well the assistant answers, so favor informative, relevant pieces: FAQs, documentation, explainers, policy pages.
Step 4: Tune the assistant
Under Hyve > Settings > Assistant you pick the AI model and the sampling behavior. The list runs from GPT-4.1 down to GPT-4o mini; more capable models cost more per message, so GPT-4o mini is the sensible starting point.
The Temperature slider controls randomness: lower values make answers more focused and deterministic, higher values make them more creative. For a support-style bot that should stick to your content, keep it low.
Step 5: Turn the chat on
Back on the Hyve > Dashboard screen, set Enable chat on all the pages to Enable. For specific pages only, use the Hyve Gutenberg block or the shortcode instead.
Now open the front end and ask something answered in a published post. If the bot quotes your own words back, the RAG pipeline works end to end.
The honest limits of the free version
Hyve Lite is genuinely usable for free, but “free” comes with four asterisks worth knowing about:
- The 500-chunk knowledge base cap. That’s roughly 150 posts averaging 1,500 words each. Bigger sites can extend it through Hyve’s Qdrant integration (Qdrant’s free plan works), while Connect’s pooled 10,000 chunks sidestep the cap entirely.
- OpenAI costs are yours on the bring-your-own-key route. Every question and answer burns a little API credit. That’s trivial on a low-traffic blog and worth monitoring on a busy store.
- No chat history or FAQ insights in free. Full chat history with export, and the FAQ Insights report showing questions the bot couldn’t answer, are Pro features. Suggested starter questions are Pro too, despite what the WordPress.org readme implies; the plugin’s own settings screen shows the upsell notice, so trust the plugin.
- It’s not self-hosted AI. Questions and retrieved chunks leave your server for processing, to OpenAI with your own key or to the hosted service with Connect. Your privacy policy should mention it.
☝️ Quick Tip: Seed the knowledge base with your 10 to 20 best FAQ-style posts instead of your whole archive. Smaller, sharper knowledge bases answer better, cost less to query, and keep you far below the chunk cap.
What else is new in Hyve 2.0 🤖
Connect aside, version 2.0 added a handful of upgrades worth knowing about:
- Lead capture (Pro). The bot can show a contact form before the chat starts, when it can’t answer, or when a visitor asks for a human. Submissions land in a Leads tab under Messages, linked to their conversation.
- Source links in answers. The widget shows a Sources badge under replies, linking visitors to the exact page an answer came from. Good for trust, and good for sending readers deeper into your content.
- Documents in the knowledge base. The bot can now learn from uploaded documents, including PDFs and Word files, so it answers from manuals, policies, or guides that never lived on a public page.
- Skills and webhooks (experimental). Skills let the bot call functions from other plugins, like looking up a WooCommerce order; webhooks stream chat activity to any URL as JSON events for Zapier, Make, or n8n. Both are early, so test before relying on them.
Bottom line 💬
RAG sounds like vendor magic until you see the moving parts: chop your content into chunks, match questions against those chunks, let the model write the answer. Hyve Lite wraps that pipeline in a free WordPress plugin, with clear trade-offs: you pay for the AI one way or another, the free knowledge base caps at 500 chunks, and the best reporting sits behind Pro.
For a blog, documentation site, or small business site with solid existing content, that’s a fair deal and a low-risk way to see what visitors actually ask. If you’d rather skip the OpenAI account altogether, Pro’s Hyve Connect handles the AI for you. Grab Hyve Lite from WordPress.org and point it at your ten best posts.
Have you tried running a chatbot on your own content? What did visitors ask that you didn’t expect?
How to Speed Up Your WordPress Site
With some simple fixes, you can reduce your loading times by even 50-80% 🚀
By entering your email above, you're subscribing to our weekly newsletter. You can change your mind at any time. We respect your inbox and privacy.
