Composum AI is a plugin that provides artificial intelligence services for Composum Pages CMS and Adobe Experience Manager (AEM), designed to help editors create, edit, and translate text content. It is based on the Apache Sling platform, so the two systems share many commonalities, and the components are kept as platform-independent as possible.
Main Features
- Translation Dialog: Can be opened on any edited text field of a multilingual website and suggests translating the component text into other languages. It can also be used for translating page descriptions, summaries, keywords, and navigation titles.
- Multi-purpose Content Creation Dialog: Helps create text for any edited text field. The text can come from a user-provided prompt, or by applying a (user-provided or predefined) ChatGPT prompt to the existing text of the page/component. It can be used to create page summaries, topic excerpts, overviews, or to request rephrasing or expansion of user-provided text snippets, etc.
- Side Panel AI: Allows chatting with it, analyzing the text content of the current page, and getting suggestions.
- Page Classification Dialog: Provides page classification suggestions (used as SEO keywords).
For the AEM version, automatic translation of a site or page tree is also implemented, transparently integrated into the rollout process of live copies. The AEM version omits the translation dialog because transparent translation is already integrated into live copies, and it also omits the page classification dialog because AEM tags work differently.
Usage
AEM version: There is a deployable package on Maven Central, and an OpenAI API key needs to be configured. For detailed installation instructions, please refer to the AEM installation guide.
Composum Pages version (currently no longer actively developed): The easiest way to try it out is through Composum Cloud, where Composum-AI is already deployed; you can get a free account, create a site, and test it. Alternatively, you can use the Composum Sling starter (a JAR that can be run locally) or the Docker image composum/featurelauncher-composum provided by the Composum Launcher project. An OpenAI API key is required and must be configured in the "GPT Chat Completion Service" on the Felix Console configuration tab.
Technical Backend
Currently, OpenAI's chat completion API is used as the backend (also available through the Microsoft Azure OpenAI service), because it is already commercially available at scale, competitively priced, and powerful. Since various open-source LLMs can run on similar interfaces, these models can also be used, for example through LM Studio or Ollama. It is recommended to use a model with performance at least equivalent to GPT 3.5. If an alternative backend is needed, you can contact the project team; usually only minor changes are required.
Optional Vision Capabilities
Optional image vision capabilities can be used, for example to generate alt text for images.
Project Intent and Development
Part of the project's intent is to evaluate the effect of using AI services (especially ChatGPT and the GitHub Copilot IntelliJ plugin) to accelerate development. The results include: the feature creation process significantly improved the quality and speed of specifying features and thinking; development acceleration is ongoing, and GitHub Copilot and ChatGPT have indeed accelerated much of the work.
More Documentation
The Composum AI homepage contains user documentation and the Maven site. The AEM architecture and Composum architecture describe the project architecture and decision records. The project structure describes the module structure. The ChatGPT API analysis contains a discussion of the ChatGPT API. Next steps contains a sketch of the planned next steps. Thanks to OPAX for providing many prompt ideas.
Limitations
The Composum Pages version is currently no longer actively developed. The project requires configuring an OpenAI API key (or a compatible local model). Specific details such as charges, pricing, and platform compatibility are not currently provided.
