Vector embeddings are becoming a key building block in GeoAI, enabling new forms of search, analysis, and data integration. But the landscape of geospatial embeddings is still emerging and often conflated, spanning foundation models, derived representations, and the systems used to manage them.

This talk provides a practical introduction to geospatial embeddings, focusing on how the pieces fit together. We will outline a simple framework that distinguishes between foundation models such as Clay, pre-computed embedding representations such as the TESSERA temporal grid, and the systems used to store and query these vectors at scale.

We will compare how these approaches differ in terms of inputs such as imagery, time series, and multimodal data, as well as training strategies and intended use cases including similarity search, classification, and change detection. We will also examine the role of general-purpose vision models and how they are being adapted for geospatial workflows.

Finally, we will discuss the current state of openness in this space, including what is truly open-source, what is partially accessible, and where gaps remain.

This session offers a structured view of the geo-embedding ecosystem, helping practitioners understand what is available today and how to evaluate and combine these approaches in real-world applications.