GMC AI Assistant for WordPress — AI drafting, scheduling and translation with approval and budget controls

Who it is for

WordPress admins and editors who want AI help drafting, rewriting, translating and scheduling content — but need approval and budget controls to stay in charge, and prefer to bring their own key rather than pay per word.

The scheduler and review queue suit teams that publish regularly and want AI to accelerate the work without taking humans out of the loop. Nothing goes live until a person approves it.

Control, not autopilot

The review queue holds AI-assisted content until a reviewer approves it.
Budget limits cap how much you can spend on any provider.
The audit log keeps a trace of what the assistant did.
Switchable providers mean you are never locked to one vendor — change when pricing or output quality changes.

What is inside

Every capability runs inside your own WordPress backend. Nothing here is customer-facing — it is an assistant for your content team.

Admin AI chat

A chatbot widget on every admin page plus a dedicated full-page chat, with configurable personalities. Draft, rewrite, translate and summarise your content in place.

Six switchable providers

OpenAI, Claude, Gemini, Groq, NVIDIA NIM and Ollama — including a local Ollama model to keep inference in-house. One automatic retry falls back on transient errors.

Scheduler + review queue

Recurring WP-Cron jobs (one-shot, daily, weekly, monthly) with row-level locking and retry with backoff. Output lands in a review queue — nothing publishes without approval.

Budget governance

Limits per user, role, provider or globally, with email alerts at 70%, 90% and 100% and a hard block at 100%. Provider spend stays predictable.

Audit log

An append-only record of AI activity with automatic PII redaction, so you always have a trace of what was generated and when.

Health check

One screen reports database schema, provider connection, crypto secret and WP-Cron status — so you know scheduled work is actually running.

Publishes into standard WordPress. Approved content is written to wp_posts via wp_insert_post(), idempotent on re-publish. Because it uses the standard call, setting the content type to “product” can route output straight to a WooCommerce product description. Ships with English, Italian and Spanish and respects your site’s multilingual setup.

Requirements

WordPress 6.4 or newer
PHP 8.2 or newer
MySQL / MariaDB with InnoDB
An API key from at least one provider (OpenAI, Claude, Gemini, Groq, NVIDIA NIM) or a local Ollama endpoint

You bring your own provider key, so there is no per-word charge from us.

Security

API keys encrypted at rest with AES-256-GCM (HKDF-SHA256 key derived from your SECURE_AUTH_KEY); the plugin refuses to load if that secret is missing
HTTPS-only AI traffic
WordPress capability gating on every admin page, REST route and form handler, with nonces on every mutation
Chatbot REST calls rate-limited to 30 messages per minute per user (configurable)
Audit redaction masks fields named api_key, token, password, secret and bearer before anything is stored

Everything included

97 real features, grouped by area — the full depth of what ships in the box.

In-admin AI chat

AI expert chat inside the backend — talk to an AI assistant without leaving the CMS admin; no separate tool, tab or subscription.
Five ready-made expert personalities — Joomla Expert, YOOtheme Pro Expert, Copywriter, PNL & Persuasion and SEO Expert, each with its own tuned system prompt and icon.
Custom personalities — create, edit, duplicate and toggle your own personas; system ones stay read-only so defaults can't be broken.
Markdown-rendered replies — responses render as formatted text (escape-first for safety), not raw plain text.
Copy any message — one-click copy-to-clipboard on every chat message.
Edit and resend — rewind your own message and the log back to that point to retry a prompt cleanly.
Persistent conversation history — chats saved per user and personality, reloaded on return, with a configurable turn limit (1–100, default 20).
Export or clear a conversation — download or wipe the current chat client-side at any time.
Site-aware answers — the assistant is fed your site name and a context description so drafts and advice fit your brand.
Honest error disclaimer — a built-in notice reminds users AI can be wrong and to verify critical information.

Chatbot placement (Joomla)

Floating chat button — a FAB anchored bottom-right or bottom-left of every admin page.
Slide-in sidebar — a docked off-canvas panel with a rotated edge toggle, RTL-aware and dark-admin-aware.
Dedicated full page — a standalone Chatbot admin screen.
Combined mode — show the floating button and the sidebar together with shared history.
Master on/off switch — enable or disable the in-admin widget site-wide.

Automated content (AI Scheduler)

Recurring content jobs — schedule the AI to generate articles, social posts, email newsletters and activity descriptions on a timer.
Flexible recurrence — once, daily, weekly or monthly, with day-of-week selection and a specific run time.
Prompt design per job — title, context, tags, extra instructions and a tone (Professional, Friendly, Adventurous, Inspiring, Informative).
Word-count control — set minimum and maximum word targets per job.
Structured output — every generation returns clean TITLE / EXCERPT / TAGS / META / BODY blocks ready for publishing.
Run now — trigger any job manually for instant output, using the same engine as the cron path.
Publish-target presets — pre-set destination category, published state, language, tags and meta description on the job.
Retry with backoff — failed jobs retry a configurable number of times before being marked failed, instead of silently dropping.
Budget-aware pausing — a job that would blow the budget is auto-paused with a notification rather than billing you repeatedly.
Pause / resume / archive — full lifecycle control over each job.
Stale-lock recovery — crashed mid-run jobs are auto-released so the scheduler never wedges.

Review & publishing workflow

Nothing auto-publishes — every generated item lands in a review queue as "pending"; a human approves before anything goes live.
Approve, reject or regenerate — reject with a reason, or regenerate with fresh instructions (e.g. "make it shorter and more direct").
Inline editing — tweak the AI's title and body before approving.
One-click publish to a real article — turn approved content into a native Joomla #__content article (WordPress: a real post), mapping category, state, language, tags and meta.
Idempotent re-publish — re-publishing the same item updates the existing article instead of creating a duplicate.
Automatic tagging — parses the job's tag list and maps it to existing site tags.
Sanitised markdown — bodies are cleaned before they become live content.

Template Library

33 ready-made templates out of the box — bilingual (IT/EN), covering landing pages, blog, SEO, email, local business, tourism, e-commerce, services and social.
Seven template types — prompt, content skeleton, layout/markup, SEO (meta + JSON-LD), email, social post and YOOtheme Pro layout JSON.
Placeholder substitution — {{key}} variables are filled in at use time.
Search and filter — full-text search on name/description/tag, plus filters by type, category, language, tone and scope.
Use, duplicate, edit, delete — apply a template, or duplicate a system one into an editable personal copy (system templates stay read-only).
Import / export JSON — move templates between sites or share bundles.
Usage tracking — each template records a usage count, last-used timestamp and a per-application log.
Joomla-compatibility hints — templates declare whether they target Articles, Custom Fields, YOOtheme or HTML.

AI providers & integrations

Six provider families, bring-your-own-key — Claude (Anthropic), OpenAI, Gemini (Google), Groq, NVIDIA NIM and Ollama; switch at any time, never locked to one vendor.
Free-tier friendly — Gemini, Groq, NVIDIA NIM and Ollama offer free/local paths; Settings links to each key page and flags which tiers are free vs paid.
Curated model menus — pick from vetted models per provider (Claude Opus/Sonnet/Haiku, GPT-4o / 4o-mini, Gemini 1.5/2.0, Llama 3.x, Mixtral, Mistral, Gemma) instead of guessing model IDs.
Local, in-house inference — point at a self-hosted Ollama endpoint to keep all inference on your own server.
Automatic fallback provider — if the primary fails on a transient error (5xx, 429, timeout, rate-limit, overload), the request retries then re-routes to a configured backup.
Smart retries — up to two retries with 250 ms / 750 ms exponential backoff before failing over.
Test-connection button — verify each provider's key and connectivity from Settings before you rely on it.
Default content language — choose the default output language (IT / EN / DE / FR).

Budget & cost governance

Token and USD caps — set a hard token ceiling and a dollar ceiling per limit.
Four scopes — apply limits globally, per user group, per user or per provider.
Daily / weekly / monthly periods — budgets reset on the cadence you choose.
Threshold alerts — configurable alerts (default 70 / 90 / 100 %) email named recipients (fallback: Super Users) once per period, deduplicated.
Three enforcement modes — hard-block at 100 %, alert-only (keep going), or downgrade.
Pre-flight cost estimation — scheduled jobs estimate token and USD cost before running and self-suspend if they'd exceed the cap.
Global spend modes — Full (all providers), Free-only (route to free-tier with a chosen fallback) or Ollama-only (force local), set once in Options.
Public pricing table — built-in USD-per-million-token pricing per provider/model powers accurate cost figures.
Reset alerts — clear fired alerts when you want the cycle to re-arm.

Roles, permissions & access control

Granular capability matrix — assign AI permissions per CMS user group across 10+ capabilities.
Capabilities include — generate, approve, manage jobs, view costs, manage templates, manage budget, publish to the CMS, manage personalities, view audit log, and manage the roles matrix itself.
Delegate without super-admin — grant a content lead full AI control without handing over site-wide administrator rights.
Safe defaults — Administrator and Super Users groups always keep every capability; their rows can't be edited by mistake.
Cost visibility gate — hide USD figures from groups that lack the "view costs" capability.

Security, privacy & GDPR

API keys encrypted at rest — AES-256-GCM with per-install random salt and HKDF key derivation; keys are never stored in plaintext.
Keys never reach the browser — provider calls run server-side only; keys are never echoed to the page or returned via REST/AJAX.
Fail-closed crypto — if encryption can't run (missing site secret), it refuses rather than silently downgrading to plaintext.
HTTPS-only AI traffic — every provider request is forced over HTTPS at both the app and libcurl layers, with certificate verification and no redirects.
Secret redaction everywhere — provider error bodies that echo a key back are scrubbed before hitting any log, notice or audit row.
Prompt-injection guard — user input is filtered for common jailbreak phrasings (EN + IT) as defence-in-depth.
Privacy consent on chat — a consent checkbox gates sending, making clear the message may be processed by the chosen AI provider.
Chat-history privacy scoping — users see only their own sessions; only audit-log-privileged admins can see everyone's.
Output escaping — consistent escaping across admin views and the chat UI; reviewed clean of XSS.
Safe-uninstall controls — optionally preserve all data on uninstall, and optionally write a gzipped SQL backup of every table before teardown.

Audit & compliance

Append-only audit log — records sensitive actions (publish, role change, budget block, crypto failures) with severity, actor, subject, context and timestamp.
Filterable audit view — page and filter the log by severity/action.
Chat-history archive — browse persisted chat sessions with message counts, tokens, provider and personality per session.
Per-call usage log — every AI call records tokens in/out, cost, duration, success/failure and (redacted) error.
Retention automation — a cleanup task prunes audit, usage, chat and alert data on configurable age windows so records don't grow unbounded.

Operations, monitoring & developer

Built-in health check — one screen verifies PHP/CMS version and extensions, database schema (all 15 tables), plugin activation, provider status and configuration sanity (including the critical site-secret check).
Provider health monitoring — a scheduled task pings each provider, recording latency and ok/degraded/down status shown in the dashboard.
Dashboard KPIs — requests, tokens, cost and pending-review counts at a glance, plus recent generated content and a setup wizard.
Setup wizard — guides first-run steps (enable plugin, configure provider, describe site, enable the task plugin).
Cron-driven jobs — run-due-jobs, provider-health and cleanup routines (Joomla Scheduled Tasks / WP-Cron) with per-run controls.
CLI secret-scrub tool — a shipped command scans and fixes any historic rows written before secret redaction existed.
Standard MVC + PSR-4 architecture — namespaced, strict_types throughout; CI lints PHP 8.2/8.3/8.4 with a PHPUnit unit suite.

Internationalisation

Three fully translated admin languages — English, Italian and Spanish across menu, dashboard, settings, budget, roles, audit log, chat history, health check, templates and the chat widget.
Multilingual content output — generate in Italian, English, German or French, or force a language per scheduled job / published article.

Trust & licensing

You own the keys and the spend — bring your own provider key; GMC charges no per-word or per-call fee, and you cap costs yourself.
Self-hosted — everything runs on your own server; no GMC middleman sees your prompts or content, and a local-Ollama path keeps inference fully in-house.
GPL licensed — GNU GPL v2 or later.
Licensed one-click updates — automatic update checks from the GMC / FuerteventuraTV gateway (~every 24 h on Joomla, ~12 h on WordPress) delivered straight to the CMS updater while your licence is active.
Download Key activation — a key emailed on purchase unlocks updates; enter it once in the licence field.
Install as one package (Joomla) — the pkg_gmcfreeai package installs the component + system plugin + task plugin together and keeps them version-aligned.
Requirements — Joomla 5.4+ (also runs on 6.x) with PHP 8.1+; WordPress 6.4+ with PHP 8.2+ and the openssl extension.

Joomla & WordPress

Near-total parity — both editions share the six providers, the ~15-table schema, the security model, personalities, scheduler, review queue, budgets, audit log, health check and IT/EN/ES translations; the Joomla side is the source of truth and WordPress mirrors it.
Joomla — administrator-only (no public frontend); ships as a package of component + system plugin (chat widget + key encryption on save) + task plugin (three cron routines); publishes approved content to #__content articles; ACL rides on Joomla user groups; chatbot can appear as floating button, sidebar, dedicated page or combined.
WordPress — a single plugin with the same workspace under an "AI Assistant" admin menu; publishes to WP posts; permissions map to WP roles/capabilities; scheduling runs on WP-Cron (custom 5-minute schedule); the widget offers floating / sidebar / both modes mirroring the Joomla placement options.

FAQ

How do I install it?
In your dashboard go to Plugins → Add New Plugin → Upload Plugin, choose the plugin .zip (keep it zipped), click Install Now and then Activate. Its settings screen appears in the admin menu after activation.
Where do I enter the Download Key?
After purchase you receive a Download Key with your licence email. Enter it on the plugin’s settings screen. The key unlocks licensed one-click updates from the GMC / FuerteventuraTV update server; keep it in place and updates keep flowing while your licence is active.
Which AI providers are supported?
OpenAI, Claude, Gemini, Groq, NVIDIA NIM and Ollama. You bring your own key and switch provider whenever you like; there is no per-word charge from us.
Will it auto-publish content?
Only if you approve it. Scheduled jobs write into a review queue and keep an approval workflow, so nothing is published automatically without a check. Provider keys stay on the server.
How do updates arrive?
Automatically. The plugin checks the GMC / FuerteventuraTV update server on its own (about every 12 hours) and shows the new version under Plugins / Dashboard → Updates while your licence is active. You can also re-download the latest build from your account at any time.
How do I confirm it’s set up correctly?
Run the health check on the settings screen (provider, API key and cron should read OK), send a test message in the admin AI chat, create a small test scheduled job and confirm it lands in the review queue, then open the audit log to see both actions recorded.
Is it available in my language?
Yes — it ships with English, Italian and Spanish out of the box and respects your site’s multilingual setup.
What does the licence cover?
The software is released under the GNU GPL. Your purchase gives you the official build, licensed one-click updates and support for one site. The WordPress licence covers the gmc-ai-assistant plugin; the Joomla edition (GMC FreeAI) is a separate product. Need it on more sites? Message us for a volume or agency arrangement.

Buy · EUR 49 Documentation All extensions

Why GMC

GPL licence

Released under the GNU GPL. You get the official build and support for one site, no obfuscation.

Updates included

Licensed one-click updates for 12 months, delivered straight through the standard WordPress updater.

Key via email

Your Download Key arrives with the licence email right after purchase — paste it in and updates flow.

Self-hosted

Runs entirely on your server with your own provider key. Inference can stay in-house with Ollama.

Buy · EUR 49 Documentation Back to all extensions

14-day refund if the product does not work as described and we cannot fix it. Support in English, Italian and Spanish.