← All add-ons

MxChat AI Search

Version 1.0.9 · Updated 5 days ago

Google-style AI Overviews for your own website's search. Visitors' queries get answered straight from your knowledge base — above your theme's normal results, as a full AI results page, or anywhere on your site via an embeddable AI search box.

Features

  • AI Overview above your default search results — works with any theme's native search bar, zero theme edits. The overview renders as clean text that inherits your site's fonts and colors.
  • Full AI results page mode — a Google-style layout with the AI answer on the left, matching article cards with thumbnails on the right, and a "Show all results" link back to your classic results.
  • Embeddable search box — drop an AI-answering search input on any page with the mxchat_ai_search shortcode or the Gutenberg block. Visitors get answers in place, no page reload.
  • Grounded answers with sources — every answer is generated from your MxChat knowledge base and links the entries it drew from. Queries below the confidence threshold show nothing rather than a made-up answer.
  • Fast by design — your search page always loads instantly; the overview streams in afterward. If the AI is ever unavailable, visitors simply see your normal results.
  • Built-in caching and per-IP rate limiting protect your API bill, with a one-click cache clear and a configurable confidence threshold.
  • Multilingual-ready — every visitor-facing string (overview heading, generating message, sources label) is customizable from settings.
  • Uses your existing MxChat provider and model — OpenAI, Claude, Gemini, Grok, DeepSeek, OpenRouter, or a custom endpoint. No separate API setup.

What's new

Version 1.0.9

  • Fixed The embeddable search box no longer goes blank when it has nothing to show. A search with no confident answer now shows a short message, and a search that fails now says search is unavailable. The Message text setting words the no-answer line, and it now always shows on the settings screen.
  • Fixed AI Overviews on Gemini models answer faster and are no longer cut off mid-sentence. They now use the same thinking level as the chatbot, with more room for the answer, and a reply that arrives in several parts or with no text is handled correctly.
  • Improved A brief outage at any AI provider is retried automatically before a search gives up, the way the chatbot does it, unless automatic retry is turned off in MxChat.
  • Fixed A Pinecone or database failure is no longer remembered for 15 minutes as a search with no answer, and Test a search now says when the knowledge base could not be read.
  • Fixed AI Search keeps working on a page served from a cache after the page's security check has expired. The page now gets a current one and runs the search again, so the visitor sees the answer instead of an error or nothing.
  • Improved A search the AI could not answer is remembered for 15 minutes, so repeating it does not cost another AI call. Changing your knowledge base or your AI Search settings takes effect straight away.
  • Fixed Re-importing a knowledge entry now replaces cached overviews. On Pinecone, a page re-imported in place still waits for the cache to expire or for Save and clear cache.
  • Improved Test a search no longer promises an overview. When a match clears the threshold it now says so, and explains that the AI still decides whether those entries answer the search.
  • Fixed On a site that reaches the internet through a proxy set in wp-config.php, AI Overviews now stream through that proxy, and behind Cloudflare each visitor now has their own search limit.
  • Fixed Security hardening and general security improvements.