Skip to content

For developers

The module is built around one extension point: the AiCKEditor plugin type. Each AI action in the editor is an AiCKEditor plugin. Add your own to put a new action in the toolbar dropdown and balloon menu.

Architecture

  • Plugin type: AiCKEditor, defined by the attribute Drupal\ai_ckeditor\Attribute\AiCKEditor and discovered from src/Plugin/AiCKEditor/.
  • Plugin manager: the plugin.manager.ai_ckeditor service (Drupal\ai_ckeditor\PluginManager\AiCKEditorPluginManager).
  • Base class: Drupal\ai_ckeditor\AiCKEditorPluginBase, which implements the shared dialog form (selected text field, generate button, editable response, save action) so a plugin only fills in the parts that differ.
  • Interface: Drupal\ai_ckeditor\PluginInterfaces\AiCKEditorPluginInterface.

Request flow

  1. The CKEditor 5 UI opens the dialog form at /api/ai-ckeditor/dialog (AiCKEditorDialogForm).
  2. The plugin's ajaxGenerate() builds the prompt and returns an AiRequestCommand, which posts to /api/ai-ckeditor/request/{editor}/{ai_ckeditor_plugin} (AiRequest controller).
  3. The controller sends the prompt to the configured AI provider through the AI module and streams the response back into the dialog's response field.
  4. Saving the dialog writes the response into the editor with an EditorDialogSave command.

Both routes require the Use AI CKEditor plugin permission (use ai ckeditor).

Entity context for other modules

When the editor is on an entity form, ai_ckeditor_form_alter() records the editing entity's type and id in drupalSettings, keyed by the form's HTML id. The AiRequest controller reads that, loads the entity, and attaches it to the chat request as metadata under entity_context. Any subscriber to the AI module's PreGenerateResponseEvent can read $event->getMetadata('entity_context') to do bundle-scoped context injection without wiring anything specific to this module.

Prompts

Default prompts are config entities under ai.ai_prompt.*. Each action stores the id of the prompt it uses in ai_ckeditor.settings (prompts.<action>). Prompt text uses placeholder tokens such as {inputText}, and some actions (Translate) render the prompt through Twig for conditional logic.

See Adding an AI action for a worked example.