LolaCore – AI Admin Agent

You: Run the weekly cleanup. Lola: Last week you decided to keep post revisions for 30 days and leave the ones on your landing pages alone. Going by that, 1,240 revisions and 1.2K expired transients are safe to clear, about 34 MB. I'll show you the exact list before anything runs. Proceed?

You didn't repeat your rules. You didn't re-explain last week. Lola already knew, because she remembers your site across sessions while every other AI tool forgets the moment you close the tab. That is the difference between a chatbot and an agent that runs your site. Running WordPress once is easy; running it every day wears you down—separate screens, repeated clicks, context gone when you close the session. That grind has a name: admin fatigue, the real cost of running a site. It hits hardest in WooCommerce, where every order and stock update adds another screen. I ran a store for ten years; what burned me out was never the selling—it was the admin. LolaCore is where that stops. Lola is an AI agent inside your wp-admin. You tell her what you need; she does the admin work and shows you exactly what she will change before anything happens. No new dashboard. No external tool. She works where you already work. She is built natively on the WordPress 7.0 AI Client and the Abilities API. She reads your database, reasons about it, and executes real actions only with your explicit approval. Full picture: documentation. Persistent memory Memory is the foundation everything else stands on. Lola remembers facts across sessions: plugin conflicts, security findings, your preferences, decisions you have made—all stored locally in your WordPress database. Important findings stay sharp; old trivia fades; pinned facts stay longer. When you start a new conversation, you don't re-explain your stack. You pick up the thread. How decay, boosting, and export/import work: persistent memory guide. 77 abilities. 10 domains. One chat.

• Site diagnostics (4 abilities). WordPress and PHP status, HTTPS, database size, security headers, SEO audit. Lola tells you what's broken before your client does.
• Plugins (8). Search WordPress.org, install, activate, deactivate, update, roll back to an earlier release, or update all. "Update all plugins" is one sentence, not 14 update screens.
• Plugin Builder (9). Describe a custom plugin in chat. Lola reads your site, co-designs a short plan, generates files with safety checks and AI code review, then you test in Playground before it touches your live site.
• Block Builder (9). Same flow for custom Gutenberg blocks: Lola scaffolds block.json, editor scripts, and the PHP render callback; you preview in Playground, then install on approval.
• Content & metadata (7). Create pages, edit posts, change status, update slugs, read and write post meta. Works with Yoast, RankMath, and other post-meta plugins.
• Users (4). Create accounts, change roles, remove users with automatic content reassignment.
• Database & cache (3). Clean post revisions, expired transients, orphaned postmeta, and spam comments. The cleanup WordPress never does for you.
• Taxonomies & media (7). Categories, tags, comments, and media files.
• Themes & settings (13). Install, update, and delete themes with safety warnings, read and write site options, manage navigation menus, review cron jobs.
• WooCommerce (13). Detected automatically. Products, orders, customers, coupons, categories, sales reports. "How much revenue this week?"—from your store data, not a dashboard.

Every ability in detail: ability reference. Store tools with examples: WooCommerce abilities. How Lola stays safe Your data never touches our servers. No telemetry, no analytics. Your conversations go from your hosting to your AI provider. That's it. Nothing happens without your approval. Every write action shows a preview card with exactly what Lola is about to do. You confirm or cancel. Every action is logged with before/after snapshots. Admin-only. She has an opinion Lola has judgment. When a change would hurt your site, she tells you before you commit and offers a safer way to get what you want. If you still want to proceed, she does it your way and records the decision, so the next time it comes up the context is already there. Any model, your choice Lola works with whatever AI provider you configure in WordPress 7.0 (Settings → AI). Agentic models perform best: GPT-5.5, DeepSeek V4, Claude 4.6 Sonnet or Opus 4.7. Gemini, Grok, Ollama, and OpenRouter work through the native connector. You pay your provider directly—no token markup from LolaCore. Each message sends only the abilities your request needs, not the full catalog. Setup: connecting your AI provider. Native to WordPress 7.0 LolaCore runs entirely on the native WordPress AI Client and Abilities API. No external HTTP libraries, no custom credential vaults. When WordPress updates its AI infrastructure, LolaCore updates with it. Spanish and Brazilian Portuguese Lola speaks your language. Set WordPress to es_ES or pt_BR and the full admin experience runs in native Spanish or Brazilian Portuguese: settings panel, chat widget, ability labels, onboarding, preview cards, errors, and status messages. More than 1,050 user-facing strings per locale — not a thin language pack. The intent router is localized for both languages. Ask in Spanish or Portuguese which pages are in draft, what needs attention today, or to list your skills — Lola routes to the right tools without English keywords. Other locales fall back to English. Lola's lane A tool that does everything does nothing well. Lola is built narrow on purpose.

• She doesn't edit theme files, core, or server config. Plugin Builder and Block Builder are the exceptions: she generates custom plugins and blocks that you own and iterate in chat.
• She doesn't see your frontend. Lola lives inside wp-admin.
• In wp-admin, she works on one site at a time. Over MCP, your editor can hold the whole fleet.

Why these boundaries exist: what Lola cannot do. Skills and playbooks: teach Lola how you work Skills are short behavioral guides in plain Markdown that shape how Lola approaches a task. LolaCore ships with 14 built in, and you can write your own or ask Lola to create one for you in chat. Skills guide. A playbook is a curated manual for one area of work: your brand guidelines, your FSE build conventions, an internal process. Upload a document (TXT, Markdown, DOCX, or HTML), Lola distills it into a draft, you review and activate it. From then on she consults it whenever that domain comes up. Playbooks guide. Free addons teach Lola your stack When you work with a specific tool every day, a free addon teaches Lola that domain—all sharing the same site memory:

• LolaCore for Elementor: manage your Elementor pages and templates from chat
• LolaCore for ACF: read and write Advanced Custom Fields in plain language
• LolaCore for SEO: audit and edit your SEO without opening another panel
• LolaCore for Forms: work with your forms and their entries from the chat
• LolaCore for Statistics: ask about your traffic and get the number, not a dashboard
• LolaCore for FSE: a block and template builder for block themes

Browse them all: lolacore.com/add-ons. For agencies and consultants Run WordPress for clients and the admin work multiplies with every site. Free addons go deep on the stack you build with: 91 for Elementor, 26 for ACF, 22 for SEO, 18 for Statistics, 15 for Forms, 17 for FSE—189 abilities on top of Lola's core. Each client site is a separate install with its own memory; contexts never bleed across clients. Preview cards and logs mean you always know what changed on a live site. LolaCore bundles a zero-config MCP server—each client site becomes a tool in Cursor, Claude Code, or any MCP client. Connect the fleet and one agent runs it: "update plugins on site A", "how did site B sell this week", "SEO audit on every site". Your agent, your rules, Lola's full toolset—abilities, memory, builders, skills, playbooks. Don't know what to say? Start here. No commands, no syntax. Three prompts you can paste right now:

"Give me a quick briefing on the site. Anything that needs attention today." "Update all plugins except Elementor and [plugin name]. Show me the list first." "How much have we sold this week? Compare it to last week."

There are 45 more in the prompt library: daily check-ins, maintenance routines, session combos.

1. Upload the lolacore folder to /wp-content/plugins/ or install via the WordPress plugin screen.
2. Activate the plugin through the "Plugins" screen in WordPress.
3. Configure your AI provider through WordPress's built-in AI settings (Settings → AI).
4. Open the floating chat widget in your admin dashboard.
5. Start managing your site through conversation.

Requirements

• WordPress 7.0 or higher
• PHP 8.0 or higher
• MySQL 5.7+ or MariaDB 10.3+ (JSON column support required)
• An AI provider configured through WordPress's AI Client settings
• WooCommerce 8.0+ (optional, for store management abilities)

Detailed setup instructions and first-conversation tips: getting started guide.

2.1.3

• Fix: MCP native abilities — lolacore/get-site-info, lolacore/get-memory-context, lolacore/get-action-history, and lolacore/get-scan-data are now exposed through the LolaCore MCP server. Previously the guard only whitelisted chat-bridge tool names (e.g. get-site-status), so MCP agents received a misleading “requires a valid license” error when calling the documented boot abilities. Uses WordPress core wp_get_abilities() (WP_Ability instances), not the non-existent wp_get_registered_abilities() helper.

2.1.2

• Feature: Deep internationalization — complete native support for Spanish (Spain) (es_ES) and Brazilian Portuguese (pt_BR). More than 1,050 user-facing strings per locale: settings UI, chat widget, ability labels, onboarding, preview cards, errors, and status messages. Shipped as compiled .mo language packs; other locales fall back to English.
• Feature: Brazilian Portuguese (pt_BR) — full translation catalog at parity with Spanish, transcreated for Brazilian product voice (você, not European Portuguese).
• Improvement: Localized intent router for Spanish and Brazilian Portuguese — natural-language keywords, meta-tool queries ("what can you do?"), holistic monitoring ("what needs my attention?"), and deferral/confirmation patterns so chat routes tools correctly in both languages.

2.1.1

• Fix: TOOL DISCIPLINE now applies even when write confirmation is disabled in Safety settings — it was previously omitted on sites with direct-write mode.
• Fix: Deferred-intent auto-continue — when Lola says she will look something up (e.g. "Let me fetch that…") but emits no tool calls, the server automatically continues the turn once and forces the read tools to run instead of waiting for the user to say "ok".
• Fix: Plugin Builder iteration install — updating a built plugin from the chat delivery card (v2 over an installed v1) now reliably prompts to overwrite the on-disk copy, completes the install, and activates afterward. Fixes a locale bug where translated admin languages never triggered the overwrite retry, and replaces the browser confirm dialog with an inline modal plus clearer activation errors on the card.

2.1.0

• Feature: Progressive tool expansion in chat — new requestAdditionalCategories meta-tool adds 1–2 domain categories to the session without loading every ability (MCP-style discover → expand → execute).
• Feature: getDiscoveredAbilities search — filter by query, category, limit, or names_only so Lola finds the right tool without dumping the full registry into context.
• Improvement: Session toolset persists across tool continuations — expanded categories are not lost mid-workflow.
• Improvement: Gemini tool cap prioritizes expanded categories and keeps expansion meta-tools available under the 32-tool limit.
• Improvement: System prompt ROUTED TOOLSET DOCTRINE — cross-domain tasks (e.g. resolve a page by title, then translate) are expected; Lola escalates before claiming a task is impossible.
• Improvement: TOOL DISCIPLINE — read lookups (search, inspect, resolve a page) must run in the same turn; no intent-only pauses waiting for the user to say "ok".

2.0.3

• Fix: Database migrator now recreates any missing LolaCore tables on load, so sites that upgraded to Playbooks without getting wp_lolacore_playbooks self-heal on the next admin request — no deactivate/reactivate required.
• Fix: Chat no longer shows a false firewall error when WordPress prints a database notice before the JSON response (for example a missing Playbooks table with WP_DEBUG on).
• Fix: Chat error messages now match the real failure (database issue, timeout, garbled response, or hosting WAF) instead of always blaming the hosting firewall.

2.0.2

• Feature: Plugin rollback for WordPress.org plugins — roll back to an earlier release from chat (rollback_plugin). Lola can list older versions first (get_plugin_versions). Premium and custom plugins are blocked with a clear message.
• Improvement: Rollback confirmation card shows a calm human preview (plugin name and version change) instead of technical fields and long version lists.

2.0.1

• Improvement: LolaCore for FSE now appears in the Addons catalog and Expansions panel before you install it, with a WordPress.org install link (21 abilities).

2.0.0

• Feature: Block Builder — describe a custom Gutenberg block in chat; Lola scaffolds block files, saves to the Built Blocks library, tests in WordPress Playground, and installs to wp-content/lolacore-blocks/ after your approval. Nine Block Builder abilities (createBlock, updateBlockCode, listBuiltBlocks, getBuiltBlock, getInstalledBlock, testBlock, installBlock, uninstallBlock, deleteBuiltBlock).
• Feature: System playbook block-development for Block Builder scaffolding and iteration.
• Improvement: Broader Spanish (Spain) coverage for builders, chat, and action handlers.
• Improvement: Playbook distiller — clearer format contract, YAML frontmatter repair, and persistent error UX when distillation fails.

1.9.0

• Feature: Playbooks — curated domain manuals Lola consults when working in a specific area. Built-in playbook-format contract, system playbook plugin-development for Plugin Builder, database and filesystem sources, model-adaptive delivery (inline hints for lightweight models, full attachment on ability use), and seven playbook abilities for chat and MCP.
• Feature: Playbooks admin under Lola → Playbooks — list, create, edit, activate, upload (TXT/MD/DOCX/HTML), AI distillation with review before save, and reload. Skills moved to its own submenu (Lola → Skills).
• Improvement: Upload guardrail with a clear notice when extracted document text exceeds 40,000 characters.

1.8.9

• Fix: Plugin Builder Playground works again on production hosting. Opening Playground no longer depends on a loopback scrape on your live server (which often fails and was blocking every library plugin with a false "fatal error"). Runtime load checks still run when saving code for review and before installing on your site.
• Improvement: Playground links are short and shareable. The blueprint is served from your site via a signed REST URL (blueprint-url), so MCP and chat agents can give you a clickable link that opens Playground in the browser without copying a megabyte-long URL from wp-admin.

1.8.8

• Improvement: The Plugin Builder now works over MCP. A connected code editor can create and iterate plugins straight into your Plugin Builder library by sending the exact code — the site no longer asks its own AI to rebuild code the external tool already wrote. The wp-admin chat is unchanged: there you describe what you want and Lola writes it. Plugins added over MCP go through the same static safety checks and must still be tested in Playground before they can be installed.
• Fix: LolaCore for Forms now appears on the Addons screen before you install it, with a one-click install from WordPress.org — the same pattern as the other free bridge addons (ACF, SEO, Statistics).

1.8.7

• Fix: On the Addons screen, installed free bridge addons now show Learn more → and link to their lolacore.com landing pages instead of Download free → (which only applies before install). Install-from-WordPress.org links are unchanged for addons you have not installed yet.
• Fix: Addons catalog and cross-domain footer links no longer 404 — marketing URLs now use UTM-safe landing pages, and How cross-domain intelligence works → points to the live add-ons catalog at lolacore.com/add-ons/.
• Fix: The View all abilities list on Settings now lists tools for free bridge addons (ACF, SEO, Statistics, Elementor, and others), not just premium addons and core domains.

1.8.6

• Fix: Resolved a race condition on sites where another active plugin bundles its own copy of the WordPress MCP adapter (recent versions of WooCommerce do this). The Abilities API init event fires only once per request; when WooCommerce or an addon triggered it before the adapter could register its meta-tools, the debug log filled with repeated "ability does not exist" notices on every wp-admin page load and Lola's IDE/MCP connection could stop working. LolaCore now registers the MCP protocol meta-abilities at the earliest hook priority and skips tools that are not yet available.

1.8.5

• Feature: The Addons screen now shows the free LolaCore bridge addons (for ACF, SEO, and Statistics) even before you install them, each with a one-click install from WordPress.org. Until now only installed addons appeared, so there was no way to discover the free ones from inside Lola.
• Docs: New conversation-led opening for the plugin description that shows how Lola uses persistent memory, plus a clearer short description and search tags. No functional changes to the agent.

1.8.4

• Docs: Rewrote the readme to explain more clearly what LolaCore does and the problem it solves (admin fatigue). No functional changes to the plugin.

1.8.3

• Fix: Improved compatibility with hosting security plugins and firewalls. Chat messages are now sent as plain text, which prevents some security suites from blocking WordPress REST API requests on affected hosts — an issue that could interfere with Application Password management and MCP connections.
• Improvement: When a hosting firewall blocks a chat message, Lola now explains what happened and what to do, instead of showing a generic connection error.

1.8.2

• Fix: Free bridge addons (e.g. LolaCore for ACF, LolaCore for Elementor) now show a green FREE ADDON badge in the What Lola Can Do ability panel, matching the premium addon label treatment.

1.8.1

• Fix: Chat now accepts HTML and JavaScript code snippets (e.g. WPCode tracking scripts with <script> tags). Messages are no longer stripped by over-aggressive sanitization, and code payloads are base64-encoded in transit so hosting WAF rules do not block admin-ajax requests.

1.8.0

• Feature: Guided onboarding, Spanish (es_ES) admin UI, Addons screen, and settings reorganization.
• Security: Confirmation-card escaping, per-admin chat session binding, and expanded protected options.

1.7.2

• Improvement: Plugin Builder pre-delivery safety checks, runtime load test, and separate load-test status on delivery cards.

1.7.1

• Fix: Remove development-only CLI repro scripts from the WordPress.org distribution package (they remain in the source repo for local debugging). Packaging-only release, no functional changes.

1.7.0

• Feature: Plugin Builder — nine abilities, Built Plugins library, Playground testing with install gate, async chat artifact cards, quality pipeline, and four built-in skills.
• Fix: Daily WP-Cron maintenance hooks now run memory decay and log cleanup.

1.6.3

• Fix/improvement: Batched database optimization (revisions, spam, orphaned postmeta, transients) with error checks, recovered-space reporting, and a time guard.

1.6.2

• Fix: Gemini compatibility for empty tool arguments and oversized toolsets.
• Improvement: Quota-aware classifier behavior for Gemini free tier.

1.6.1

• Fix: MCP filters scoped to the LolaCore server, external tool response formatting, and a PHP 8.1 deprecation in the MCP adapter.

1.6.0

• Feature: Bundled zero-config MCP server, IDE parity, cross-environment memory, and dynamic tool syncing.

1.5.0

• Feature: Theme management (five abilities, full lifecycle parity with plugins).
• UX Fix: Clearer post-action summaries and idempotent confirmation cards.

1.4.1

• Fix: Orphan tool_call_id errors during multi-tool runs.
• Improvement: Four-layer tool-call pairing defenses and sanitizer telemetry.

1.4.0

• Feature: Skills.MD — markdown skills with triggers, admin panel, conversational authoring, and schema 0.3.0.
• Improvement: listSkills ground-truth rule and session-summary locale respect.

1.3.0

• Performance: Deferred memory extraction and faster DeepSeek classifier calls.
• Fix: Meta-tool pattern encoding and holistic intent limited to the current message.

1.2.0

• Feature: Memory Bootstrap from addons and holistic monitoring across domains.
• Fix/improvement: Tool-call pairing fixes, DeepSeek reasoning handling, and ~30–40% token savings.

1.1.3

• Architecture: Memory persistence overhaul — pinned facts, importance-aware decay, proactive directives, and session summaries.
• Compliance: i18n prep, error sanitization, dependency updates, and dev-file removal.

1.1.2

• Architecture: ActionExecutor split into 12 domain handlers; getDiscoveredAbilities and manageRoles.
• Fix/improvement: Intent-router ordering, plugin search, chat autolinks, centralized ErrorMessages, and chat error UX.

1.1.1

• Fix: Taxonomies and featured image on create/edit content. Feature: uploadMedia.

1.1.0

• UI redesign, memory maintenance from settings, autonomous tool escalation, SafetyConfig, and addon infrastructure.

1.0.2

• Fix: Preserve HTML in chat. Enhancement: Markdown rendering with marked.js and DOMPurify.

1.0.1

• Filters for external addon actions and SOP map injection.

1.0.0

• Initial public release — 51 abilities, persistent memory, site scanner, WooCommerce tools, MCP support, and preview/confirm/execute safety flow.