Attendant – AI Site Search & Content Finder
AI chat search that answers visitor questions from your own WordPress content
WordPress's built-in search is a blunt keyword match against post_title and post_content. It has no idea what to do with a visitor typing "apartments in Lisbon under 500K with 3 bedrooms", that's a filtered query, not a keyword search, and no amount of stemming fixes that. Attendant treats site search as a chat problem instead: a visitor asks a question in plain language, the plugin decides whether it's a structured search over your listings or a general knowledge question about your content, and it answers inline with clickable buttons linking straight to the matching pages.
The interesting part is how it avoids guessing. Structured queries go through OpenAI's function-calling (tools) API, which forces the model to return real parameters, price, bedrooms, location, instead of a text blob that then has to be parsed and hoped to be right. Those parameters run as a normal WP_Query straight against the WordPress database, not an HTTP crawl of the site. Schema discovery is automatic too: at index time the plugin scans registered post types, taxonomies, and custom fields, including ACF, MetaBox, and WooCommerce attributes, so there's no manual field mapping to maintain. Open-ended questions fall back to a RAG path, chunked embeddings matched by cosine similarity with source citations, which is optional and off by default since the 2.0 rebuild; a set of admin-defined Q&A pairs is checked first via semantic matching before either path runs.
It's aimed at sites where content doesn't fit into a simple menu: real estate directories, job boards, WooCommerce stores, event listings, medical directories, anything built on custom post types. It's bring-your-own OpenAI key with no SaaS layer in between, and the privacy defaults are deliberately conservative, conversation logging and lead capture both ship off, IP addresses are never stored, and session identifiers are one-way hashed.
What it does
Dual search modes
Routes each visitor question to structured Smart Search over WP_Query or RAG-based Knowledge Q&A, picking the right mode automatically.
Auto schema discovery
Scans post types, taxonomies, and custom fields from ACF, MetaBox, and WooCommerce to build a searchable schema with no manual config.
Function-calling search
Uses OpenAI's tools API to extract structured search parameters from natural language instead of parsing free text.
Direct database queries
Runs searches straight against WordPress via WP_Query rather than crawling pages over HTTP, for faster and more accurate matches.
Guided no-results flow
When a search comes up empty, the assistant offers alternatives drawn from the site's own content and asks one clarifying question at a time.
Privacy-first defaults
Conversation logging and lead capture ship off by default, IP addresses are never stored, and session identifiers are one-way hashed.
Built with
FAQ
Do I need an OpenAI account?
Yes. You need an OpenAI API key from platform.openai.com. The plugin uses your key directly — we never see it.
How much will it cost to run?
With gpt-4o-mini (our recommended model), each conversation turn costs approximately $0.005 USD. 1,000 conversations per month ≈ $5–10. Content indexing for a 500-page site costs approximately $0.025 (one-time).
Is my data sent to OpenAI?
Only the user’s chat message and relevant content excerpts (for Q&A mode) are sent to OpenAI. Your full database is never sent. API keys are encrypted before storage and never leave your server.
Will it slow down my site?
No. The chat widget JS is less than 10KB gzipped and loads asynchronously. Admin scripts only load on plugin admin pages. All AI processing happens via REST API calls triggered by the visitor — never on page load.
Does it work with ACF, MetaBox, WooCommerce?
Yes. The schema discovery engine automatically detects custom fields from ACF, MetaBox, and WooCommerce product attributes.