Abstract

Visual exploration of embedding spaces allows researchers to discover similarities, neighborhoods, clusters, gradients, and shifts across complex datasets. In practice, this usually relies on a patchwork of computational tools that can be difficult to set up, maintain, or even use without a graphical interface. EIDORA is an open-source desktop research software that integrates the main stages of exploratory embedding space analysis in a local, project-based environment. It supports importing and linking data sources, computing vector embeddings, running projection and clustering algorithms, exploring two-dimensional projections through navigation, filtering, and visual encoding, and producing publication-ready figures and animations. EIDORA supports multimodal data together with rich contextual metadata, as commonly found in cultural collections. Its modular architecture accommodates new models and algorithms as they emerge, while preserving data, parameters, and visual states as reusable research artifacts for reproducible analysis.