The Drupal AI module gives Drupal a common layer for artificial intelligence. Instead of each feature integrating its own vendor, you configure providers once and every AI capability on the site uses them. For organizations, that means choice, control and a consistent place to govern AI.
The architecture in one picture
- Core AI module: defines operation types (chat, embeddings, text-to-image, speech, translation and more) and a common API for calling them.
- Provider modules: connect to hosted or self-hosted model providers. You can configure several and pick a default per operation.
- Key module: stores API credentials outside of configuration exports.
- Feature submodules: editor assistants, automators, search, translation, logging and assistants that build on the API.
Installation follows the usual Composer workflow:
composer require drupal/ai drupal/key
# plus a provider module for the model service you choose
drush en ai key -y
# then enable the features you need, for example:
drush en ai_ckeditor ai_automators ai_logging -y
Five workflows worth implementing first
1. Editor assistance in CKEditor
Give editors in-place tools to summarize, adjust tone, fix grammar or translate a selection. The editor stays in control: suggestions are applied only when accepted.
2. Automated fields with Automators
Automators fill a field from other fields when content is saved: generate a meta description from the body, suggest taxonomy tags, extract structured data from an uploaded PDF, or create alt text for images. Configure them per field, with your own prompts.
3. Semantic search
AI Search plugs into Search API and stores embeddings in a vector database, so visitors can search by meaning, not just keywords. It is also the retrieval layer for RAG-style assistants that answer from your published content.
4. Translation support
For multilingual estates, AI translation can produce first drafts of translations that human translators then review, a significant time saver for Arabic and English sites.
5. Assistants and agents
Assistants can answer site-specific questions, and the companion AI Agents project explores agents that help site builders perform configuration tasks. Treat these as powerful tools that need clear permissions.
Governance inside Drupal
Because every call goes through one layer, Drupal becomes a natural control point:
- Permissions: decide which roles can use which AI features.
- Logging: record prompts and responses for review and audit.
- Provider choice per use case: route sensitive content to a provider or self-hosted model that meets your data residency rules.
- Human review: combine automators with content moderation so AI output is never published without approval.
Practical tips
- Start with one high-volume, low-risk task such as alt text or meta descriptions.
- Version your prompts in configuration and review them like code.
- Measure editor time saved and acceptance rate of suggestions.
- Keep the AI module and providers updated; the ecosystem moves quickly.
See how this fits Drupal's direction in our Drupal 12 AI roadmap article, or talk to our Drupal team about a pilot.