Kjeks AI Reviewer

Suggests a category and metadata for each unreviewed tracker using WordPress 7's core AI client. Advisory only — you confirm every suggestion.

Requires WordPress 7.0+PHP 8.3+Requires the Kjeks core Source on GitHub →

101 · Getting started 201 · How suggestions work 301 · REST, cron & internals

101 Getting started

Reviewing a large tracker inventory by hand is slow. The AI Reviewer proposes a category and metadata for each unreviewed tracker so triage becomes a quick yes/no instead of research.

Requirements

WordPress 7.0+ — it uses the core AI client (wp_ai_client_prompt).
The Kjeks core plugin, active. Without it the add-on only shows an admin notice.

Install & activate

Download the latest kjeks-ai-reviewer.zip and upload it under Plugins → Add New → Upload Plugin, then activate it (network-activate on multisite). It adds no menu of its own — instead it appears as a tab on the existing Cookie Consent screen. Updates arrive automatically from GitHub releases via plugin-update-checker.

Run a review

Open Cookie Consent and switch to the AI Reviewer tab.
Trigger a suggestion pass for the unreviewed trackers.
For each suggestion, read the rationale and confidence, then Accept or Reject.

Accepting a suggestion is what marks a tracker reviewed and applies the category. Nothing is changed until you confirm it.

201 How suggestions work

Database first, AI second

The reviewer first looks the tracker up in a bundled copy of the Open Cookie Database. A confident, complete match avoids an AI call entirely. Only when that misses does it prompt the AI model.

What each suggestion contains

Field	Notes
`category`	One of `necessary`, `preferences`, `analytics`, `marketing`.
`party`	`first` or `third`.
`provider`, `purpose`, `retention`	Enriched metadata for the cookie declaration.
`documentation_url`	Reference link, when known.
`confidence`	How sure the model is.
`rationale`	Short explanation you can sanity-check.
`model`, `grounded_by`, `generated_at`	Provenance: which model, what it was grounded on, and when.

How the category is chosen

The reviewer is deliberately conservative: it picks necessary only when an item is clearly essential (session, auth, security), and prefers marketing when unsure. Any label it can't map to a Kjeks category also falls back to marketing — never necessary. So genuinely third-party resources (e.g. Google Fonts, Gravatar) usually land on marketing — a safe default for a consent tool, not a bug.

You keep the final say: Accept takes an optional category override, so you can reclassify a suggestion (or self-host the resource to make it first-party) before confirming.

Advisory, not authoritative. The model can be wrong. Suggestions never set reviewed on their own — an administrator accepts each one.

301 REST, cron & internals

The AI call

The reviewer asks the core client for a strict JSON answer with a small token budget:

wp_ai_client_prompt( $prompt )
    ->using_max_tokens( 400 )
    ->generate_text();

REST API

All routes live under kjeks-ai/v1 and require manage_network on multisite, or manage_options on a single site.

Method & route	Body
`GET /state`	Current pending trackers and their suggestions.
`POST /suggest`	Optional boolean `force` to regenerate.
`POST /settings`	Required boolean `cron_enabled`.
`POST /accept`	Required `id`, optional `category` override.
`POST /reject`	Required `id`.

POST /wp-json/kjeks-ai/v1/accept
{
  "id": "_ga",
  "category": "analytics"   // optional override of the suggestion
}

Weekly cron (opt-in)

A scheduled job, kjeks_ai_weekly_suggest, can pre-generate suggestions for new trackers. It is disabled by default — enable it by posting cron_enabled: true to /settings (stored in kjeks_ai_settings).

Options

Option	Holds
`kjeks_ai_suggestions`	Suggestions keyed by tracker ID.
`kjeks_ai_settings`	Add-on settings (currently `cron_enabled`).

Admin integration

The tab is injected client-side: the add-on enqueues on toplevel_page_kjeks-network and registers itself through the kjeks.networkAdminTabs JavaScript filter — no separate admin page.