mad.Design AI Sales and Support Chatbot

Most chatbots answer based on general knowledge. Yours doesn’t have to.

Design AI Sales and Support Chatbot searches your own pages, posts and products first. Then it answers with the passages it actually found, so every reply is grounded in your content.

It’s built for 2 jobs:

Pre-sales questions. A visitor wondering about delivery times, sizes or pricing gets an answer on the spot, with a link to the page that covers it. No contact form, no waiting hours for a reply, no lost sale in between.

Customer support. Questions about returns, setup or opening hours get answered from your own help pages, so the same five questions stop landing in your inbox.

When a conversation needs a person, the chatbot asks for an email address and sends you the full conversation. You follow up knowing exactly what was asked and what the visitor has already read.

Your key, your costs. Connect an API key from OpenAI (ChatGPT), Anthropic (Claude), Google (Gemini) or any OpenAI-compatible provider. You pay the provider directly at their rates, with no markup and no extra subscription.

Answers come from your content

• Indexes the post types you choose into a vector store on your own server, either as HNSW index files or in a native MariaDB VECTOR column.

• Hybrid retrieval: semantic search and keyword full-text search run together and the rankings are merged. The right page is found whether the visitor uses your wording or their own.

• Contextual retrieval: every chunk gets a one-line summary at index time that names its page and section, so short passages are still found.

• Page context: attach notes to a single page that the chatbot sees whenever a visitor is on that page.

• Page builders are indexed with their rendered output, not raw shortcodes: Elementor, Bricks, Divi, Beaver Builder, Brizy, Thrive Architect, SiteOrigin and WPBakery, with auto-detection per post.

Searches only when needed

A greeting does not need a knowledge base lookup. A product question does. The assistant decides and calls the search itself, so simple messages take one round-trip and real questions get a full retrieval.

Tools the assistant can call:

• search_knowledge_base: semantic and keyword search across your indexed content

• query_posts, get_post, get_page_content: look up live posts and pages

• list_post_types, get_taxonomy_terms, get_posts_by_term: browse your site structure

Each tool can be switched off. Developers can register their own through a filter.

Cost and speed

• Prompt caching: the fixed part of the prompt stays identical between turns, so providers that support caching charge less and answer faster.

• Long conversations are condensed into a running summary, so the prompt does not grow with every message.

• Replies are generated in a background job. A slow provider never blocks your page.

• Performance logging shows the time spent per stage of each reply.

Privacy and data control

• Optional GDPR consent overlay before the first message.

• Email addresses and phone numbers are masked before a message reaches the AI provider. The originals stay in your database for follow-up.

• Indexed content and embeddings stay on your server. Only the passages needed for the current question are sent.

• Per-IP rate limiting and retention periods for conversations and logs.

• API keys can be stored in wp-config.php instead of the database.

• Conversations are readable in the admin, with optional email notification.

External services

This plugin sends data to third-party AI services to generate replies and to create the vector embeddings used for retrieval. No requests are sent until you configure a provider and API key.

• The visitor’s message, the conversation history, the system prompt and any retrieved site content are sent to the chat provider to generate a reply.

• Text you choose to index is sent to the embeddings endpoint configured under Settings → Knowledge. This endpoint has its own URL, model and API key. It defaults to OpenAI (https://api.openai.com/v1/embeddings) and accepts any OpenAI-compatible endpoint, including one you host yourself. That service’s terms and privacy policy apply.

Chat providers:

• OpenAI: https://api.openai.com. Terms: https://openai.com/policies/terms-of-use/ Privacy: https://openai.com/policies/privacy-policy/

• Anthropic (Claude): https://api.anthropic.com. Terms: https://www.anthropic.com/legal/commercial-terms Privacy: https://www.anthropic.com/legal/privacy

• Google (Gemini): https://generativelanguage.googleapis.com. Terms: https://ai.google.dev/gemini-api/terms Privacy: https://policies.google.com/privacy

• Custom endpoint: the base URL you enter under Settings → AI Engine. That provider’s terms and privacy policy apply.

If you configure “Remote Instructions URLs” or a “Static context URL”, the plugin also fetches those URLs to build context.

Source Code

assets/dist/ contains built files. Sources ship with the plugin: `assets/js/frontend/`, `assets/js/admin/`, `assets/css/`.

js 20+: npm install, then npm run build (Tailwind CSS CLI + esbuild, both MIT). json.