MxChat Admin Assistant

A ChatGPT-style AI chat that lives inside your WordPress admin. Ask questions about your store, your SEO, or anything else — and get real answers without opening a second tab.

Overview

MxChat Admin Assistant turns your WordPress dashboard into a working AI workspace. You get a familiar chat interface — sidebar with thread history, model picker, image-gen toggle, web-search toggle — but plugged directly into your site. Ask WooCommerce questions in plain English and the AI queries your real orders, products, and customers. Connect Google Search Console (via the Advanced Content Editor add-on) and the same chat can surface content gaps, quick-win keywords, and underperforming pages. Underneath, you pick from 100+ AI models — Claude, GPT-5, Gemini, Grok, Llama, DeepSeek, Mistral — using your existing MxChat API keys (no extra setup) or an optional OpenRouter key for the full catalog.

This is the admin counterpart to MxChat's public-facing chatbot. It's for site owners and operators who want their site data and an LLM in the same window.

Setup

1. Install MxChat Admin Assistant via Plugins → Add New → Upload Plugin (upload the ZIP) or from the MxChat add-ons gallery.
2. Activate via Plugins → Installed Plugins.
3. Make sure the MxChat core plugin is active and your MxChat Pro license is activated under MxChat → Pro & Extensions.
4. Open the assistant at MxChat → Admin AI Chat (or Network Admin → Settings → Admin AI Chat on multisite).
5. Pick a model from the model dropdown and start chatting. Your existing MxChat API keys are used automatically.

Settings

Model selection

Click the model picker in the chat toolbar to choose any built-in model — OpenAI (GPT-5.6 and the rest of the GPT-5 line, GPT-4.x), Anthropic (Claude Haiku 5.5, Sonnet 5.5, Opus 5.5, Fable 5.1 and earlier Claude 5 and 4.x models), Google (Gemini 4 Argon, Gemini 3.x, 2.5), xAI (Grok 4.7 and earlier Grok models), and DeepSeek (DeepSeek Flash, V4 Flash and V4 Pro) — or any OpenRouter model you've loaded. Each new chat can use a different model — the choice is saved per thread. If you pick a model whose provider has no API key saved in MxChat, an inline hint appears by the picker naming exactly which provider key to add under MxChat → Settings — so you catch a keyless model before a send fails, not after.

Thread persistence

Chats auto-save after every message. The sidebar lists your 10 most recent threads — click any to resume right where you left off. Refreshing the page restores your current chat.

Where the history is stored (3.1.10). Saved conversations now live in two dedicated database tables, {prefix}mxchat_admin_chat_threads and {prefix}mxchat_admin_chat_messages, instead of one ever-growing entry in the WordPress user profile. That entry used to be loaded on every page view of the site, admin or not — WordPress loads a user's whole profile meta whenever any plugin looks that user up — and on an install with a long history it cost about half a second and most of the request's memory, on every request. Existing conversations are moved across automatically the first time the WordPress admin loads after the update, in batches, and a compressed backup of the old data is written first to wp-content/uploads/mxchat-admin-chat/backup-<user>-<time>-<token>.mxbak (gzip-compressed JSON, protected from direct download); the profile entries are removed only once every thread is in the tables. Long conversations open faster too: the Assistant loads the latest 50 messages and offers a Load earlier messages button at the top for the rest, and each new message is stored on its own instead of the whole conversation being rewritten on every save.

Chat History (3.1.10)

A Chat History section in the Assistant settings lets you choose how many days to keep conversations (default 90, 1–3650) and clear your own history in one click. Older conversations are tidied once a day, and also when you visit the admin if scheduled tasks are not running on your site. The retention sweep does not run until the migration above has finished and the backup exists.

Prompt library

Prompts you use repeatedly can be saved and reused. The Prompts button sits in the tool row beside the composer; clicking it opens a compact panel directly above the input containing a search box, your saved prompts, and a save row for whatever is currently typed. Click a prompt to drop it into the composer. Hover a prompt to rename or delete it. Typing a forward slash as the first character in an empty composer opens the same panel, so you can reach a saved prompt without leaving the keyboard.

Image generation

Toggle the image-gen button in the chat toolbar to switch into image mode. Pick the exact model per request from the Image Model dropdown:

• Google Gemini: Nano Banana 2.1 (latest, better text and multi-turn edits), Nano Banana 2 (4K, up to 14 reference images), Nano Banana 2 Lite (fastest, lowest cost), Nano Banana Pro (top quality, 4K)
• OpenAI: GPT Image 2.5 Flare (latest, fast everyday images), GPT Image 2.5 Sunburst (highest quality), GPT Image 2 (4K, high text accuracy)
• xAI Grok: Grok Imagine Image 2.0 (latest), Grok Imagine Image Pro, Grok Imagine Image

You can also paste or upload an image to edit.

From 3.1.12, DALL-E 3 is no longer offered and a saved DALL-E 3 selection moves to GPT Image 2.5 Flare automatically. The Nano Banana preview models move to their released versions. GPT Image 1 and the original Nano Banana (Gemini 2.5 Flash Image) are no longer offered for new picks; a site still using one keeps it until the provider's retirement date (GPT Image 1: October 23, 2026), then moves to the replacement automatically. A notice tells you when a model was switched.

From 3.1.13 the Image Model list follows MxChat's own image model list, so models MxChat adds or retires show up here without an Admin Assistant update (the list above is what MxChat offers at the time of writing). Your saved model is unchanged. The setting always shows the model your images are actually made with. Older OpenAI image models a site may still have saved (GPT Image 1.5, GPT Image 1 Mini and ChatGPT Image Latest) generate with OpenAI, as picked, until December 1, 2026, then move to GPT Image 2.5 Flare. Image options follow the model you pick and are kept on save, including each provider's own choices when you switch between models, and image editing is offered with the Nano Banana models.

WooCommerce integration

Loads automatically when WooCommerce is active — no configuration needed. The AI can query revenue, top sellers, order counts, customer spend, average order value, and stock levels. To disable, deactivate WooCommerce (or just don't ask shop questions).

Google Search Console integration

Requires the MxChat Advanced Content Editor add-on with GSC connected. Click the chart-icon GSC toggle in the chat toolbar to enter SEO mode. The toggle is mutually exclusive with WooCommerce mode (one at a time). Alt+G is the keyboard shortcut.

Web search (Perplexity)

Toggle the Perplexity icon in the toolbar to ground replies in real-time web results. Useful for "what changed today" questions and current events.

OpenRouter API key

Optional. Under Settings, paste your OpenRouter key, click Load Models, pick one, and save. This unlocks 100+ models beyond the built-in list.

Features

• WooCommerce Analytics Chat: Ask plain-English questions like "What was my revenue this month?", "Show me my top 5 sellers", or "Which products are low in stock?" — the AI queries your live store data and returns real numbers, not guesses.
• Google Search Console Chat: Surface content gaps (high impressions, low clicks), quick-win keywords near page 1, page-specific keyword breakdowns, and cannibalization analysis. Page URLs are matched to your exact WordPress posts so the AI references real post titles.
• 100+ AI Models via OpenRouter: Switch between Claude, GPT-5, Gemini, Llama, Mistral, DeepSeek, Grok, and more — per chat. Built-in models work out of the box with your MxChat keys.
• Persistent Chat History: Sidebar shows your 10 most recent threads with smooth animations. Auto-saves after every message, restores on refresh. From 3.1.10 history lives in its own tables with a retention setting and a Clear history button, and long threads page 50 messages at a time (see Thread persistence).
• Image Generation & Editing: Generate with Gemini Nano Banana 2.1, Nano Banana 2 (4K, up to 14 reference images), Nano Banana 2 Lite, Nano Banana Pro, GPT Image 2.5 Flare or Sunburst, GPT Image 2, or Grok Imagine Image 2.0. Edit by uploading or pasting an image.
• Web Search Toggle: One-click Perplexity integration brings live web results into any conversation.
• Lives Where You Work: No second dashboard, no second login. Dedicated Admin AI Chat menu inside WP admin.
• Multisite Aware: On multisite networks, the assistant is available under Network Admin → Settings.
• Streaming replies with a Stop button (3.1.8+): Answers appear word by word as they are generated instead of arriving all at once after a wait. A Stop button is shown while the Assistant is replying — click it to end generation early and keep the partial answer, which is saved to the conversation like any other reply.
• Prompt library (3.1.8+): Save the prompts you use every day and insert them with one click from a panel anchored to the composer, or by typing a forward slash in an empty input.
• Honest answer limits (3.1.7+): If an answer ever hits the provider's length limit, the Assistant says so with a clear notice and offers to continue, rather than presenting a half-finished answer as complete. Works across OpenAI, Claude, Gemini, xAI, DeepSeek, and OpenRouter. The same release stopped OpenRouter answers getting silently cut off mid-sentence — reasoning models like DeepSeek spend part of their allowance on internal thinking, so the Assistant now requests a proper response budget from every provider, and Gemini answers are no longer capped to a few short paragraphs.
• Discontinued models move automatically (3.1.12+): When a provider discontinues a model, every user's Admin Assistant model and the image model move to the current replacement, with a notice saying what changed.
• Model support tracks core (3.1.7+): The Assistant reads which models accept which settings from the main MxChat plugin rather than keeping its own copy, so models added to MxChat keep working here without a separate add-on update. Vision requests follow the same path — image analysis always uses the current recommended vision model for your provider.

API & Integration

• WooCommerce: Auto-detects WooCommerce and loads a dedicated analytics handler (MxChat_Admin_Chat_WooCommerce) that translates natural-language questions into store queries (orders, products, customers, revenue, stock).
• Google Search Console: Routes GSC queries through the Advanced Content Editor add-on's connection. Supports overview, page performance, keyword performance, content gaps, quick wins, page+keyword pairs (cannibalization), and per-page keyword breakdowns.
• Image Generation: Concrete model routing — settings auto-migrate from legacy "Gemini" / "Grok" labels to the specific model (e.g. nano-banana-2). From 3.1.13 the offered models come from core MxChat's image model list (core 3.2.27 or later; older cores use the add-on's built-in list). Uses MxChat's stored API keys for OpenAI, Gemini, and xAI.
• Perplexity: Bundled endpoint (MxChat_Perplexity_Endpoint) for real-time web search; uses your MxChat-configured Perplexity key.
• OpenRouter: Optional integration — paste a key in Settings to expand the model dropdown to 100+ models. Models are cached for fast loading.

Requirements

• MxChat (mxchat-basic): Required and active. Provides the API key storage and core plumbing. Keep core current (3.2.26 at the time of writing); the Assistant takes its model rules from it.
• MxChat Pro License: Required — activate at MxChat → Pro & Extensions. See pricing.
• WordPress: 5.8 or newer.
• PHP: 7.4 or newer.
• WooCommerce: Optional. Required only for store analytics chat.
• MxChat Advanced Content Editor: Optional. Required only for Google Search Console chat (and GSC must be connected inside that add-on).
• OpenRouter API key: Optional. Required only to unlock the extended 100+ model catalog.

Troubleshooting

"MxChat Admin Chat requires the MxChat plugin to be installed and active"

The core MxChat plugin isn't active. Install and activate mxchat-basic under Plugins → Installed Plugins.

"MxChat Admin Chat requires MxChat Pro to be activated"

The Pro license is missing or inactive. Go to MxChat → Pro & Extensions (or admin.php?page=mxchat-activation) and activate your Pro key. The Admin Assistant menu won't appear until the license is active.

GSC toggle does nothing / says prerequisites missing

The Google Search Console chat needs the MxChat Advanced Content Editor add-on installed with GSC actually connected. A modal will guide you through both steps the first time you click the toggle.

WooCommerce questions return generic answers

The WooCommerce handler only loads when WooCommerce is active at the moment the assistant initializes. Make sure WooCommerce is active, then refresh the admin page.

Image generation produces wrong style or fails on existing settings

Settings from older versions auto-migrate to the equivalent concrete model on upgrade. Before 3.1.13, an older OpenAI image model (GPT Image 1.5, GPT Image 1 Mini, ChatGPT Image Latest) could be sent to Gemini instead, and image options could be lost on save; update to 3.1.13. If something still looks off, reopen the image-gen model dropdown and re-pick your preferred model.

Perplexity web search stopped working after saving settings

Update to 3.1.13. Earlier versions could replace your Perplexity model with a discontinued one when you saved the settings, which broke web search. 3.1.12 fixed most cases and 3.1.13 stops saving from sending a discontinued Perplexity model.

GPT-6 replies are slow

Update to 3.1.12. Earlier versions sent GPT-6 models OpenAI's default reasoning setting, which is slower; 3.1.12 sends each GPT-6 model the setting it needs.

The Assistant errors out on a specific model

Several model-specific failures were fixed in 3.1.7 — update the add-on first. That release fixed the Assistant failing on Claude Sonnet 5, on the GPT-5.6 models, and on GPT-5 Vision for image analysis, and restored DeepSeek after DeepSeek retired its older API models on 24 July 2026. If your site still has a retired DeepSeek model saved, the picker now offers DeepSeek V4 Flash and V4 Pro and switches the old selection to V4 Flash automatically. Because the Assistant now takes its model rules from core MxChat, keeping the core plugin current is the fix for most "works on one model, not another" reports.

The site got slower the longer you used the Assistant

Update to 3.1.10. Earlier versions stored every conversation in the WordPress user profile, which WordPress loads in full on every request that touches that user — so a long history was paid for on every page view, by every visitor. 3.1.10 moves it into dedicated tables automatically, backup first.

Can I undo the chat history migration?

Yes. Before 3.1.10 moves your conversations out of the user profile it writes a compressed backup to wp-content/uploads/mxchat-admin-chat/backup-<user>-<time>-<token>.mxbak. To roll back: decompress the file, take its threads and conversation values, and store them again as the mxchat_admin_chat_threads and mxchat_admin_conversation user meta for that user (they are PHP-serialized values, the same shape 3.1.9 stored), then reinstall 3.1.9. Nothing in 3.1.10 deletes the backup, and the new history tables are left untouched by a rollback, so you can move forward again later.

An answer stopped mid-sentence

From 3.1.7 the Assistant tells you when it hits the length limit and offers to continue, so a partial answer is never presented as finished. If you're seeing silent cut-offs, you're on an older version — update. Note this is separate from the "conversation has reached the maximum allowed length" notice, which is about the whole thread; there the fix is to save the chat and start a new thread.

Last reviewed by Sage on 2026-10-25. Spotted an issue? Open a ticket and we'll patch the doc.