Embedding Atlas

Embedding Atlas is an MIT-licensed toolkit for working with large text or image embeddings. It provides interactive, in-browser visualizations with cross-filtering and metadata search, and ships as both a Python package (CLI tool, Jupyter widget, Streamlit component) and an npm UI component library. It computes embeddings and 2D projections locally (data stays on-device), and includes WebAssembly implementations of UMAP, approximate nearest-neighbor search, and density-based clustering. Integrations include SentenceTransformers for embedding generation and optional loading of datasets from Hugging Face.