Embedding Atlas
A tool for interactively visualizing large embeddings and their metadata, letting a user cross-filter, search and explore data structure at scale.
For embeddings, Embedding Atlas provides automatic data clustering and labeling, kernel density estimation with density contours, order-independent transparency for overlapping points, real-time nearest-neighbor search, and WebGPU-powered rendering up to a few million points. For any tabular data, it adds linked dashboards and cross-filtering with standard chart types such as bar, line, bubble, count plot and eCDF and a composable chart spec, built-in viewers for text, image, audio, numeric, categorical and time columns, and AI agent access through MCP to query the schema, run SQL, create charts and capture screenshots.
It installs with pip install embedding-atlas and runs as a command-line tool or a Jupyter notebook widget through EmbeddingAtlasWidget; its components are also published as an npm package for use in plain JavaScript, React or Svelte.
It is written in TypeScript and licensed under MIT.