AIForum.

How to Add a Local AI Chatbot to WordPress Without Third-Party APIs

On this page

THE SHORT ANSWER

You can add a working chatbot to WordPress without sending visitor messages to OpenAI, Anthropic, or any hosted service. Two documented self-hosted approaches exist: rule-based keyword matching with JRMora Local Chat Assistant, and retrieval-style search with BM25 ranking in MBR Intelligent Site Assistant. Both run entirely on your own server, need no API keys, and carry no per-query billing.
Quick facts

A local chatbot can be added to WordPress without routing a single visitor message to OpenAI, Anthropic, or any other hosted service. Two documented, self-hosted approaches are available: rule-based keyword matching, as implemented by the JRMora Local Chat Assistant plugin, and retrieval-style search using BM25 ranking, as implemented by the MBR Intelligent Site Assistant. Both run entirely on your own WordPress server, require no external API keys, and avoid per-query billing.

This guide is written for site owners, creators, and developers who want immediate visitor assistance or FAQ responses while keeping interactions on their own infrastructure. It explains what “local” means in this context, lists prerequisites, walks through configuration for both plugin types, describes how to verify the setup, and outlines the tradeoffs that should drive your choice.

What “Local” Means for a WordPress Chatbot

In this context, “local” does not necessarily mean a large language model running on your web server. Neither of the plugins described here ships model weights or performs neural inference. Instead, they handle visitor language with deterministic techniques that run inside PHP and JavaScript on the same machine that serves your site.

The JRMora plugin is explicit about this: it is “a lightweight, high-performance local chatbot plugin that provides automated instant responses to visitor queries using simple keyword matching rules,” and it operates “100% locally on your WordPress installation” with “no external API keys, no tracking, and no external requests.” The MBR project takes a different route, describing itself as “a WordPress chat widget that answers visitor questions in plain English using your own published content — no language model, no third-party API, no per-query billing,” built from “pure PHP, BM25 ranking, intent matching, synonym expansion, and a conversational UI scoped entirely to your server.”

So the phrase “local AI” here refers to a self-contained matching and ranking system, not a generative model. That distinction matters for both expectations and hardware: you do not need a GPU, and you do not need to install a runtime such as Ollama or llama.cpp.

Two Practical Approaches

Keyword-rule chatbots

A keyword-rule chatbot maps visitor phrases to pre-written answers. The site owner defines a set of tags (keywords or phrases) and a corresponding response for each. When a visitor’s message contains a matching tag, the bot returns the associated answer. This is predictable, fully under editorial control, and cheap to run.

The JRMora plugin implements exactly this model. Its admin panel lets you “define keyword rules (tags) and corresponding responses (res)” with client-side pagination and instant keyword filtering. It also supports plain-text export and import of the entire knowledge base, automatic conversion of plain http:// and https:// URLs inside responses into clickable links, and a REST API logging endpoint that records visitor queries for admin review.

Self-hosted retrieval chatbots

A retrieval chatbot searches your existing published content instead of relying on hand-written rules. The MBR project uses BM25, the ranking function it notes is also used by Wikipedia, over title, excerpt, and body fields. It adds a Porter stemmer so that “running” matches “ran,” synonym groups so that “shop” finds “store,” and intent matching so that common questions can return pre-written answers without a search at all. The result is a conversational widget that answers from your own content rather than from a rules table you maintain by hand.

Prerequisites

  • A working WordPress installation with administrator access to the plugin screen.
  • The ability to upload and activate a plugin (self-hosted WordPress.org installs support this directly).
  • For the JRMora approach: a written list of the questions visitors actually ask, and the answers you want the bot to give.
  • For the MBR approach: enough published content for the ranking to have something to search, since it answers from your own posts and pages.
  • No GPU, no model download, and no external API account are required for either option.

Option 1: JRMora Local Chat Assistant (Keyword Rules)

  1. In your WordPress admin, go to Plugins → Add New, search for “JRMora Local Chat Assistant,” install it, and activate it.
  2. Open the plugin’s admin panel. This is where the knowledge base of tags and responses is managed.
  3. Add your first rule: enter a keyword or phrase as the tag, and the response the bot should return when that tag is matched. Repeat for each FAQ or common query you want covered.
  4. Use the plain-text export to review or bulk-edit the full knowledge base outside WordPress. You can paste a large set of rules back in through the import function rather than clicking through the interface one rule at a time.
  5. Place the shortcode [jlca_chatbot_local] on the page or post where the chatbot should appear. According to the plugin description, its CSS and JavaScript are registered but only enqueued on pages where this shortcode is actually used, so other pages are not loaded down with chatbot assets.
  6. Check the admin dashboard and log management view to see recent query logs and to track answered versus unanswered questions. Unanswered entries are the clearest signal of which tags you still need to add.

Expected result: visiting the page with the shortcode should render a chat widget, and sending a message containing one of your tags should return the paired response. If a response includes a plain URL, it should appear as a clickable link.

The MBR project is distributed from its GitHub repository rather than the WordPress.org plugin directory, so installation follows the standard manual route for a self-hosted plugin.

  1. Download the plugin from the repository at github.com/HarbourBob/MBR-Intelligent-Site-Assistant.
  2. In WordPress, go to Plugins → Add New → Upload Plugin, upload the package, and activate it.
  3. Confirm that your site has published content for the ranking to search. Because the bot answers from your own posts and pages, an empty or very thin site will produce weak results.
  4. Review any configuration options the plugin exposes for synonym groups and intent answers. Adding synonyms for terms your audience uses informally — for example, mapping “WP” to “WordPress” — improves match quality.
  5. Add the chat widget to the pages where you want it, following the plugin’s own placement instructions.

Expected result: asking a question in plain English should return a ranked list of relevant content from your site, or a pre-written intent answer for common questions. The project states that no visitor data leaves your server.

Verifying the Setup

Verification should cover three things: whether the widget renders, whether answers are correct, and whether anything is leaving your server.

  • Rendering: load the page in a private browser window and confirm the widget appears. For JRMora, also confirm that pages without the shortcode do not load the chatbot’s CSS or JavaScript.
  • Answer quality: send the exact questions your visitors ask, plus paraphrases. For rule-based bots, a paraphrase that lacks your tag will not match; for BM25 bots, stemming and synonyms should absorb some variation.
  • Privacy: open your browser’s network panel and watch for requests to external domains while chatting. Both plugins claim no external requests, and this is the simplest way to check that claim on your own installation.
  • Logging: for JRMora, review the query log and the answered-versus-unanswered breakdown to find gaps in your knowledge base.

Limitations and Tradeoffs

Neither approach is a general-purpose conversational agent, and it is worth being clear about that before you commit.

Keyword-rule chatbots only answer what you have written rules for. They are fast, predictable, and fully auditable, but they will not handle unexpected phrasing unless you keep expanding the tag list. The JRMora plugin’s own logging system exists precisely because unanswered queries are expected and need review.

BM25 retrieval chatbots handle more natural phrasing because they rank your existing content rather than matching exact keywords. But the quality of the answer depends on the quality of your published content, and the bot presents retrieved passages rather than generating new prose. It also requires ongoing content maintenance: thin or outdated pages will surface as poor answers.

Both approaches avoid the recurring costs, external dependencies, and privacy questions that come with hosted AI APIs. The tradeoff is capability: you are not getting the open-ended reasoning of a large language model. For FAQ handling, documentation search, and contextual help, that is often exactly what a site needs. For open-ended conversation, it is not.

Choosing Between Them

Choose the keyword-rule approach when your visitor questions are finite and known, when you want complete editorial control over every response, and when you want the smallest possible footprint on page load. Choose the retrieval approach when you have a substantial body of published content and want the bot to answer from it without you writing a rule for every phrasing.

It is also reasonable to combine the ideas: use intent matching for the handful of questions that deserve a hand-written answer, and let retrieval handle the long tail.

Next Steps

Start small. Install one plugin, add five to ten rules or confirm that your content is indexed, and watch the logs for a week. The unanswered queries are your roadmap. If you run into a specific error, an unexpected match, or a widget that fails to render, share your WordPress version, PHP version, plugin version, and the exact message in the Ai-Forum.org community so others can help reproduce it.

FAQ

Do I need a GPU or a local language model to run these chatbots?

No. Neither plugin ships model weights or performs neural inference. JRMora uses keyword matching rules, and MBR uses BM25 ranking over your published content, both running in PHP and JavaScript on your existing WordPress server.

How is a local chatbot different from a hosted AI chatbot?

A hosted AI chatbot sends each visitor message to an external provider and typically bills per query. The plugins described here process interactions on your own server with no external API keys and no per-query cost, but they do not offer the open-ended reasoning of a large language model.

Can I back up or bulk-edit the chatbot's answers?

The JRMora plugin supports plain-text export and import of its entire knowledge base, which lets you back it up, edit it outside WordPress, or paste a large rule set in at once.

How do I find out which questions the bot cannot answer?

JRMora includes a REST API logging system and an admin dashboard that shows recent query logs with an answered-versus-unanswered breakdown. Unanswered entries indicate which tags or rules you still need to add.

References

  1. https://wordpress.org/plugins/jrmora-local-chat-assistant/
  2. https://github.com/HarbourBob/MBR-Intelligent-Site-Assistant

Related terms

BM25
A ranking function used in information retrieval that scores documents by term frequency and rarity; the MBR plugin uses it to rank your published content.
Keyword matching
A rule-based technique where a visitor message is compared against a list of tags, and a pre-written response is returned when a tag matches.
Porter stemmer
An algorithm that reduces words to a common stem so that variants such as "running" and "ran" can be treated as related during search.
Shortcode
A WordPress syntax enclosed in square brackets that inserts dynamic content into a post or page; JRMora uses one to place the chat widget.

Leave a Reply

Your email address will not be published. Required fields are marked *