AI-Driven Semantic Search in Cesium Sandcastle Now Available

In December 2025, we shared our progress on AI-driven semantic search for Cesium Sandcastle. This feature is now generally available. 

Sandcastle has long been a cornerstone of the CesiumJS developer community: showcasing new features, sharing ideas, learning, prototyping, experimenting, and validating code. Its accessibility means developers can quickly find, build, and share code examples relevant to their projects, with demonstrations on how to implement features in the CesiumJS library. 

With the semantic search, users are now able to make a wider range of searches to more efficiently find sandcastles relevant to their queries. You no longer need to exactly match the name, description, or code to find applicable examples. 

This is powered by an open-source embedding model, which embeds the user's query and compares it to the semantic embeddings that are now precomputed for the full code contents for all sandcastles. This open-source embedding model is run locally on the end user's hardware, so your search data never leaves your device, just as before. By leveraging cutting-edge models optimized for small size, search performance remains fast. 

The semantic search functions as a hybrid with the previous search, so any existing search results will still appear as before. To opt out of the semantic search, you can disable the "Semantic Search" toggle in settings. 

Try the new semantic search here, and share feedback on GitHub.