AI engineering
Editorial AI, grounded search and the structured data that makes your content citable, run inside your own cloud account with a person reviewing before anything publishes.
We put language models to work inside Drupal on the jobs they do well: alt text, summaries, tagging, and answers from your own pages with citations. They run in your tenancy, your content trains no one else's model, and nothing publishes without a person saying so.
Talk about AI on your platform- Models run in your tenancy: Bedrock, Azure OpenAI or Vertex
- Nothing publishes without human review
- Structured data and llms.txt so engines quote you correctly
What AI is for on a content platform
Most of the value of a language model on an institutional site is unglamorous: alt text for ten thousand images, summaries and tags for an archive nobody has time to read, a first draft of a plain-language version, a search that answers in sentences from your own pages and says so when it does not know. The Drupal AI module gives all of that a home inside the CMS, with more than ninety-five providers behind one interface, so the choice of model is a configuration, not a rebuild.
What we do with it
We design and build the editorial AI features your team will actually use, with the review step in the workflow rather than bolted on. We build site-grounded search and assistants that answer only from your content, with citations to the page. We connect the models through your own cloud account, Amazon Bedrock, Azure OpenAI or Google Vertex, so prompts and content stay inside your tenancy and your procurement rules. And we do the outward-facing half: schema.org structured data, clean canonical pages and an llms.txt file, so that when an AI answer engine cites a source on your subject, it cites you and gets it right.
How we work
Every feature starts with the question it answers and the person who reviews the result. We measure before rollout: how often the draft is accepted, how often the search answer is right, how often it correctly declines. Pilots run on a slice of content with named reviewers, and only a measured pilot becomes a rollout. Costs are metered per feature in your own cloud bill, and every prompt and output is logged so a wrong answer can be traced.
Why now
The Drupal AI module reached 1.4.8 in September 2026 and runs on Drupal 10.5 and 11.2 or later; Drupal CMS 2.0 ships optional AI tools with governance. AI answer engines now sit between your content and the people looking for it, and they cite the pages they can parse. The organizations that publish structured, canonical, plainly written pages are the ones being quoted. That is engineering work, and it is the same work that makes a site accessible and findable.
The state of AI in Drupal
1.4.8
Drupal AI module, current release
Released September 2, 2026. Runs on Drupal 10.5 and 11.2 or later.
Drupal AI on drupal.org95+
Providers behind one interface
Anthropic, AWS Bedrock, Azure, Google Vertex, OpenAI and more. The model is a setting.
Drupal AI on drupal.org100%
Of outputs reviewed before publishing
Our rule for every editorial AI feature we build, not a target.
0
Prompts or content leaving your tenancy
Models run in your cloud account, under your keys and your procurement rules.
Our cloud engineering serviceI want to …
add AI to the editorial workflow
Alt text, summaries, tags, plain-language drafts and translations, generated inside Drupal and reviewed by the editor before they save.
- The three or four jobs your editors would hand off first, chosen with them
- Drafts land in the form as suggestions, never as published content
- Acceptance rate measured, so the feature earns its keep
answer questions from my own content
Search that replies in sentences with citations to your pages, and declines when the answer is not in them, for staff, students, patients or the public.
- Your content indexed as embeddings in your own cloud account
- Every answer cites the page it came from, and says when it cannot
- Right, wrong and declined answers measured in a pilot before rollout
appear correctly in AI answers
When an AI answer engine covers your subject, it should cite your page and quote it accurately. That is structured data, canonical pages and an llms.txt file.
- schema.org on every page type, checked on every build
- An llms.txt file that says what you publish and where
- Answers that cite you, checked monthly
tag and describe a decade of content
An archive with no alt text, no summaries and inconsistent tags can be described in weeks with a model, and checked by people at the rate they can manage.
- A pilot on a sample, with the error rate measured
- Batches sized to the reviewers, with a queue they control
- The alt text pass is also an accessibility remediation
keep AI inside my cloud account
Procurement and privacy rules often forbid sending content to a public API. Bedrock, Azure OpenAI and Vertex put the same models inside your own tenancy.
- Provider chosen for your cloud, keys and billing under your account
- Prompts and outputs logged, so a wrong answer can be traced
- Cost metered per feature on your own bill
know whether an AI feature is worth it
Before building, a short pilot on real content and real reviewers says whether the model is good enough for the job and what it costs per month.
- Two weeks, one feature, named reviewers
- Accuracy, acceptance and cost measured, in writing
- A go or no-go you can defend
An AI feature, from pilot to rollout
Every feature goes through the same gate. The automated steps run on every build; the human ones decide whether it ships.
A feature that does not clear the pilot does not ship. That is most of the value of running one.
What is changing in AI on Drupal
Checked on drupal.org on September 10, 2026.
The Drupal AI module
Now the standard foundation: one API over ninety-five-plus providers, with vector search, automators, assistants and agents inside the CMS. Version 1.4.8 shipped on September 2, 2026 for Drupal 10.5 and 11.2 or later.
Drupal CMS 2.0
Ships optional AI tools with governance, which is the right framing: the tooling is there, and the decisions about what it may touch belong to the organization.
Private model endpoints
Bedrock, Azure OpenAI and Vertex put the same models inside your own tenancy. The procurement objection to AI is mostly gone; the content stays in your account.
AI answer engines
A new front door to your content that favors pages which are structured, canonical and plainly written. The work that passes an accessibility audit is the work that gets you cited.
Services that pair with this one
AI on a platform is also a cloud account, a content model and an accessibility pass. These are the services most often bought alongside.
-
Cloud engineering
We design and run the cloud your platform lives on, in your own account, as code, with monitoring, patching and a monthly cost review in writing. We do not resell hosting, and everything we build is documented so the next engineer can run it.
-
Drupal development
New builds, major-version upgrades, module and theme work, and migrations from Drupal 7 onward. Every site we build is one we can keep patched, accessible and on a supported version for years, because that is the job after launch.
-
Accessibility & AI visibility
Accessibility fixed at the source, in the theme and the editorial rules, with a conformance report you can hand to procurement. The same work makes your pages the ones AI engines cite.
Tell us the job.
The task you would hand to a model first, who would review it, and where your content is allowed to go. We come back with what a pilot would look like and what it would cost.
Talk about AI on your platform