Semantic Content Indexing

When the feature flag LPS-122920 is enabled, the platform can now create text embeddings (numerical representations of input text, also known as vector embeddings) for Object entries. These embeddings are generated using the content from the searchable fields and are meant to capture the meaning and the context of the content. You can select from various available third-party providers and models, such as OpenAI or Hugging Face, to generate these vectors.

Customizable Search

Now that semantic indexing is also supported for Object entries, you can create highly customized searches using Blueprints. Specifically, the Rescore by Text Embedding query element is available for use.This element automatically handles the process of creating vectors from the user’s keywords through the configured provider.

For instance, this capability allows you to combine traditional keyword search with AI-powered vector search techniques to implement hybrid search, now also for Object entries. This combination is quickly becoming the new standard for modern content search and discovery.

Key Benefits:

More Relevant Search Results: Create search experiences that understand the meaning and context of your Object based content or application data and user searches.

The Semantic Search capability is planned to be moved to GA status in early 2026. This timeline is intended to outline Liferay’s general product direction and it is subject to change.

In this case, the feature “Enable JavaScript Client Extensions for Administration in SaaS Environments” has remained under Release Feature Flag for three consecutive quarters. During this period, it has demonstrated strong stability, with no reported bugs or user complaints.

Following this evaluation, we’ve decided to move the feature to GA, based on its proven reliability, adoption, and positive feedback.

Style Books now support 8-digit hex colour codes (e.g. #RRGGBBAA), which include an alpha channel to define opacity. Previously, these values only worked when set as defaults in the frontend-token-definition.json file; manual input in the editor would strip the opacity digits. With this update, customers can directly enter and save full 8-digit hex values in the Style Book editor, and the opacity is correctly reflected in the UI.

Key Benefits:

• User will be able to define colours with both RGB and transparency in a single, standardised format improving Liferay’s design flexibility.
• Aligns with modern CSS and browser support for 8-digit hex codes.
• No need for more workaround formats (like rgba) to manage opacity.

From now on most CSS files in Liferay have hashed file names at build time. For example, a clay.css file may appear at run-time with a randomly generated hash value in the name, like clay.(tvERyCVfuRc).css.

This hash value represents a unique version of this file, so the browser can identify that the file's contents haven't changed. This allows the file to remain in Liferay's cache indefinitely.

For those files that can not be hashed, because they are generated in runtime by the server depending on some parameter such as the css tokens, a new onfiguration is available in DXP to configure the TTL and the possibility to add the no-cache header, that ensures the revalidation of the asset with the server before to being served.

Also, hashed files have a fallback strategy based in TTL+eTag if they are called by their canonical name, this is a fallback for error in the import maps or old portlets that doesn’t know the name of a hashed file.

Key Benefits:

 

The new Liferay DXP caching strategy for CSS files improves performance and stability.

1. Faster Page Loads: Significant reduction in subsequent load times.

2. Elimination of Stale Resources: Hashed URLs prevents users from seeing outdated CSS after an update.

3. Reduced Origin Server Load: Less server overhead as browsers retrieve unchanged files directly from their local cache, saving CPU and bandwidth.

4. Cache busting: Updated resources automatically force the browser to fetch the new version, but updating their file name with a new hash when content changes.