Every AI chatbot plugin promises answers drawn from your content instead of generic internet noise. Few explain how that works, and fewer still let you try it for free. This guide does both: a plain-English explanation of retrieval-augmented generation (RAG) and a step-by-step setup of Hyve Lite, a free WordPress AI chatbot that answers visitors using your posts and pages.
A quick disclosure: Hyve is our plugin, built by Themeisle (the company behind WPShout), so you’ll get straight talk about limitations and what the free tier won’t do.
What RAG actually is
RAG stands for retrieval-augmented generation. It’s the technique that makes chatbots appear “trained on your content” without actually fine-tuning a large model on your site. No training runs, no custom model, no long GPU jobs. In three simple steps:
1) The question becomes a number. When a visitor asks something, the chatbot converts the question into an embedding — a numeric fingerprint that captures meaning.
2) Your content is searched by meaning. Your posts and pages are split into small chunks and already converted into the same kind of embeddings. The bot finds the chunks whose embeddings are closest to the question.
3) The AI writes a grounded answer. The best-matching chunks are sent to the language model along with the original question, and the model composes a reply that’s grounded in your actual text.
Two practical consequences follow. The good: answers come from what you published, so update a post and resync to change future responses. The cautionary: the bot is only as good as your content — short, thin, or outdated posts lead to short, thin, or incorrect answers.
Setting up a WordPress AI chatbot on your content
Overall time: about 15 minutes. Most of that is choosing which posts and pages to include in the knowledge base.
Step 1 — Install Hyve Lite
In your WordPress dashboard go to Plugins > Add New, search for “Hyve”, then install and activate Hyve Lite. Activation adds a Hyve menu in the sidebar where you’ll manage everything — no coding required.
Step 2 — Choose how the AI is accessed: your OpenAI key or Hyve Connect (Pro)
If you want to use your own OpenAI account, go to Hyve > Settings > Advanced, paste your OpenAI API key into the API Key field, and save. Visitor messages then go directly from your site to OpenAI and consume credits from your account.
If you have a Hyve Pro license, you can use Hyve Connect instead. That removes the need to paste an API key — Hyve Connect handles the OpenAI access for you and shows pooled usage meters. With Hyve Connect the OpenAI key fields remain inactive.
Step 3 — Build the knowledge base
Open Hyve > Knowledge Base and click Add Data. You’ll see your posts and pages listed with an Add button next to each. Pick the content that should power the answers.
Don’t just add everything. The selection determines how well the assistant answers. Prefer helpful, detailed pages: FAQs, documentation, product guides, policy pages, or long explainers. Excluding irrelevant or low-quality posts improves response accuracy.
Step 4 — Tune the assistant
Under Hyve > Settings > Assistant choose the model and sampling behavior. Options range from higher-capability models (costlier) down to lightweight models like GPT-4o mini. For most support-style use cases, a smaller model with sensible cost is the right start.
Use a low Temperature to keep answers focused and deterministic. Higher Temperature increases creativity but can make replies stray from your source content — not ideal if you want factual, content-grounded answers.
Step 5 — Turn the chat on
From Hyve > Dashboard set Enable chat on all the pages to Enable, or add the Hyve Gutenberg block or a shortcode where you want the chat to appear. Then open your front end and ask a question covered by one of your published posts. If the bot quotes or closely matches your text, the RAG pipeline is working.
Practical limits and trade-offs
– You are still paying for AI: either via your OpenAI credits when using your key or indirectly via the Hyve Connect plan if using Pro.
– The free knowledge base has a chunk limit (currently 500 chunks). Large sites or exhaustive knowledge bases may need Pro or to curate the content carefully.
– Advanced reporting and usage controls are available in Pro.
– Answers reflect the quality and coverage of your content. Good documentation produces good answers; sparse content produces sparse answers.
Bottom line
RAG is straightforward: split your content into chunks, turn both questions and chunks into embeddings, retrieve the most relevant chunks, and let the model generate an answer that’s grounded in that content. Hyve Lite packages that pipeline into a free WordPress plugin and offers a low-risk way to test an on-site chatbot. For blogs, documentation sites, or small business sites with solid existing content, it’s an effective, inexpensive experiment.
If you don’t want to manage an OpenAI account, Hyve Pro’s Hyve Connect handles the AI for you. Try Hyve Lite from WordPress.org and point it at your ten best posts to see what visitors actually ask.
Have you ever run a chatbot on your content? What unexpected questions did visitors ask?