Drupal AI 1.2.0-alpha1 was just released and it comes as a feature preview of what is coming with the AI 1.2.0 release.
Note that since this is an alpha, we will not provide upgrade paths from this alpha and more features will still be added before the beta releases.
Article by Marcus Johansson.
Field Widget Actions - invoke AI from anywhere
The new Field Widget Actions module makes it possible to add interactive buttons on any entity form, to make it easy for editors to interact and fill out fields with AI from anywhere. With a push of a button you can have suggestions or picks - if you are not happy, you can push and ask again. This ensures that we can use AI to help the editor, but the editor has the final decision.
This works together with the AI Content Suggestions, AI Automators and AI Agents module, to have anything from simple preset configurations to custom logic business or even agents taking autonomous decisions on how to help fill out a field.
Possible solutions that come out of the box, but are not limited too, are picking categories, creating tags, helping with writing summaries, coming up with titles, extracting e-mails from sources.
In conjunction with the AI Automators, there are at least 100+ different use cases for these buttons. With agents, you can have the agent go out and research things on the Internet for you, to give you suggestions on improved Meta tags for instance.
And for advanced site builders, you can invoke this via ECA, which already has an integration. That together with the option to create your own processor makes the possibilities endless for this.
Prompt Library
With the release of 1.2.0, we make it possible to ship prompts as configurations. This will also make it easier for third party modules to give suggested prompts to use for their products.
Since context engineering is an important part of a well working AI ecosystem, and since these can be hard to understand, we now have the possibility to ship them together with any module.
Gone will be the times when you would have to guess or ask a maintainer for best practice prompts, we make them possible to ship.
This also means that you can make your own business stand out, by using well-crafted prompts, that you can share over projects via recipes.
AI Content Suggestions on Block and Taxonomy Terms
For the longest time the content suggestions were only allowed on Nodes, we have now also added the possibility to add them to Block and Taxonomy Terms entities.
Mocking library for testing or replaying requests
One of the big issues with developing AI, is that you spend a lot of time waiting for AI providers to process the result. It can also be expensive, if you need to run your query over and over.
Another issue is when you write automated testing, that you need to replicate how an AI provider responds, without necessarily needing to send requests to an AI provider for similar reasons.
We have now added the possibility to mock any request done to a real provider and replay it while developing or while testing.
Allow more file types for AI models
The LLM models are evolving, and the vision models are not the only files that can be ingested anymore. There are models that can read PDF files or even look at videos. We have changed the foundation of the AI core module to make it possible for providers to feed them with any type of files.
New Automator Types
The AI Automators make it possible to set up AI in a simple manner, how a field should be automatically filled using AI. With the addition of the Field Widget Actions, these will also be available with a user interface for the editors. It already has 60+ features, the following is added as of 1.2.0-alpha1:
- Image Alt Text - this can look at any image, and use different context fields, like the article, the filename of the image and more to generate a proper alt text for the image. This will work magic for accessibility.
- Image Filename Rewrite - in a similar manner for SEO, it is important that your image files are named expressively, however when you load them directly from a digital camera, they might be named DSC_0034.jpg. The AI can now look at them after you uploaded them, and rename them to something that actually describes what is in the image.
- Summary Generation for Text with Summary - if you use Text with Summary field, this automator can look at the text and generate a summary for it, in the summary field.
Vector databases are abstracted for recipes
All vector databases have some base functionality that all of them share. This means that from a feature point of view it doesn’t really matter if you use Milvus, Postgres or Pinecone.
However in recipes you would need to specify exactly that, meaning that a recipe that adds a “chat-with-your-documents” feature, would only work with one service.
We had already abstracted this away for AI Providers, and with 1.2.0-alpha1 it will be possible to ship a whole AI and vector database solution that doesn’t care about what AI providers or vector databases you have, it just works.
Versioned Documentation
With the speed of how fast AI is moving and the AI module in its turn is moving, versioned documentation is something that is needed to make sure that you can locate the latest documentation for the version you are working with.
Our documentation is now not only versioned, but does actually version based on the major or minor version you push changes to the documentation for.
Helper abstraction for providers
If you are a provider contributor or are developing a provider in-house for private reasons and that provider is based on the OpenAI standard that many providers support - Mistral, Anthropic, Ollama and more - we now have an abstracted base class that you can use, making it a lot easier to get started with a contributed or custom provider.
This will also make it easier for us to do incremental updates and improvements on these providers, without the person developing providers needing to change their provider.
Thank you to the following contributors for helping out:
scott_euser, gxleano, marcus_johansson, breidert, kevinquillen, mrdalesmith, prashant.c, annmarysruthy, arthur_lorenz, dan2k3k4, matthews, svendecabooter, sarvjeetsingh, anjaliprasannan, valthebald, apmsooner, leo pitt, davidlfg, michaellander, andrewbelcher, narendrar, arisha, emacoti, efpapado, a.dmitriiev, akhil babu, bisonbleu, kristen pol, prabha1997, sundflux, arkener, rajab natshah, jan kellermann, emma horrell, rszrama, g.rocchini, doxigo, adwivedi008, ankitsingh0188, shalini_jha, sayyedhali, lukasfischer, yautja_cetanu, jurgenhaas, lussoluca, sijumpk, arwillame
Thank you to the following organizations for helping out:
Factorial GmbH, FreelyGive, 1xINTERNET, Velir, QED42, Drupal India Association, Soapbox, Acquia, drunomics, Amazee Labs, amazee.io, Sven Decabooter, EntityOne, Make It Fly, Dropsolid, SystemSeed, Globant, Elevated Third, Zoocha, Common Sense Media, utdanning.no, Salsa Digital, INDICIA, Vardot, werk21, The University of Edinburgh, Centarro, Ramsalt Lab, Cognizant Technology Solutions, LakeDrops, SparkFabrik, Calibrate