AI Valve

Control, meter, and permission-gate AI usage from plugins that connect through the WordPress 7 AI connector.

Inspired by WordPress AI Connectors Need More Friction, Not Less. Works with WordPress 7 RC2. Tested with WordPress AI, Virtual Media Folders AI Organizer, and AI Provider for Azure AI Foundry.

Features

• Per-plugin access control — Allow or deny individual plugins from making AI requests.
• Token budgets — Set daily and monthly token limits per plugin and globally.
• Context restrictions — Control which execution contexts (admin, frontend, cron, REST, AJAX, CLI) may trigger AI calls.
• Usage dashboard — Token consumption at a glance: summary cards, progress bars, and per-plugin breakdowns.
• Request logging — Every AI request is logged with provider, model, capability, tokens, and caller attribution.
• Budget alerts — Admin notices and optional email when usage approaches or exceeds limits.
• Developer hooks — Filter and action hooks to extend behaviour without modifying the plugin.

Requirements

• WordPress 7.0+
• PHP 8.3+
• A configured AI provider in Settings → Connectors

Installation

1. Download soderlind-aivalve.zip
2. Upload via Plugins → Add New → Upload Plugin
3. Activate via WordPress Admin → Plugins
4. Configure settings via Settings → AI Valve

From source

git clone https://github.com/soderlind/ai-valve.git
cd soderlind-aivalve
composer install
npm install && npm run build

Usage

After activation, AI Valve intercepts all calls made through wp_ai_client_prompt(). Navigate to Settings → AI Valve:

• Dashboard — Token usage for today and this month, per-plugin access/budget controls, provider breakdown, and recent requests.
• Settings — Master switch, default policy, context restrictions, global budgets, and alert configuration.
• Logs — Filterable, paginated request log with purge controls.

How It Works

AI Valve hooks into three WordPress 7 AI connector events:

Hook	Purpose
wp_ai_client_prevent_prompt	Gate requests — evaluate policy
wp_ai_client_before_generate_result	Insert a pending log row with caller attribution
wp_ai_client_after_generate_result	Update the pending row with token usage and status

A pending log row is created before the AI provider is called. If the provider throws (auth error, timeout, bad deployment), a shutdown handler marks the row as error so failed requests are never lost.

Caller attribution uses debug_backtrace() to identify which plugin initiated the request.

When a request is blocked the calling plugin receives a WP_Error with code prompt_prevented. See docs/how-blocking-works.md for the full explanation.

Developer Hooks

AI Valve exposes hooks so you can extend its behaviour from another plugin or functions.php without editing the source.

Hook	Type	Purpose
soderlind_aivalve_plugin_policy	filter	Override the allow/deny decision for any plugin
soderlind_aivalve_request_denied	action	React when a request is blocked
soderlind_aivalve_request_completed	action	React when a request succeeds (token counts available)

See docs/hooks.md for signatures, parameter descriptions, and examples.

FAQ

How do I block all plugins and only allow specific ones?

1. Go to Settings → AI Valve → Settings.
2. Set the Default policy to Deny.
3. Switch to the Dashboard tab.
4. In the Per-plugin access table, set each plugin you want to permit to Allow.

Everything not explicitly allowed will be denied.

Can I override the policy programmatically?

Yes — use the soderlind_aivalve_plugin_policy filter. See docs/hooks.md.

Development

Tests

# PHP (PHPUnit 11 + Brain Monkey)
composer install
vendor/bin/phpunit

# JavaScript (Vitest)
npm install
npx vitest run

Project structure

soderlind-aivalve.php                    Plugin bootstrap and metadata
uninstall.php                   Cleanup on uninstall
readme.txt                      WordPress.org readme
README.md                       GitHub documentation
CHANGELOG.md                    Release notes
.wordpress-org/                 WordPress.org icons, banners, screenshots
docs/                           Developer documentation
  README.md                     Documentation index
  hooks.md                      Filters/actions reference
  how-blocking-works.md         Blocking flow details
src/
  Plugin.php                    Hook registration orchestrator
  Admin/AdminPage.php           Settings page shell for the React admin app
  Alert/AlertManager.php        Budget threshold notices and email alerts
  Interceptor/
    RequestInterceptor.php      WP 7 AI hook wiring and pending-row logging
    PolicyEngine.php            Allow/deny/context/budget evaluation
    CallerDetector.php          Backtrace to plugin slug attribution
  REST/UsageController.php      REST API endpoints for admin screens
  Settings/Settings.php         Options read/write/sanitize
  Tracking/
    LogRepository.php           Custom DB table CRUD and migrations
    UsageClock.php              Database-aligned date buckets
    UsageTracker.php            Rolling daily/monthly token counters
  js/
    index.js                    React admin entry point
    App.jsx                     Admin app shell and tab routing
    api.js                      REST client wiring
    admin-api.js                Shared JS API helpers
    admin.css                   Admin UI styles
    components/                 Dashboard, settings, logs, and tables
build/                          Generated admin assets
tests/
  Unit/                         PHPUnit + Brain Monkey tests
  js/                           Vitest tests
  stubs/                        Test stubs for WordPress AI connector
composer.json                   PHP dependency metadata
package.json                    npm scripts and JS dependencies
phpunit.xml.dist                PHPUnit configuration
vitest.config.mjs               Vitest configuration

Documentation

See docs/README.md for a full index.

License

GPL-2.0-or-later — see LICENSE.

📦 Source: soderlind/ai-valve · Edit on GitHub