Repolode
Submit
apple

apple/embedding-atlas

Interactive visualization tool for exploring large embedding datasets along with their metadata, with cross-filtering and search.

View on GitHubWebsite
Stars
4,990
Forks
332
Open issues
34
Last commit
4 hours ago

What is it?

Embedding Atlas is a visualization tool for large sets of embeddings and the metadata attached to them. It lets you explore the overall structure of your data, see dense regions and outliers, and search for points similar to a query or an existing item.

For embeddings, it provides automatic clustering with labels, kernel density estimation with density contours, order-independent transparency for overlapping points, and nearest-neighbor search. Rendering uses WebGPU and is aimed at datasets of up to a few million points. For any tabular data, it offers linked dashboards with bar, line, bubble, count, and eCDF charts, a composable chart spec for custom charts, and built-in viewers for text, image, audio, numeric, categorical, and time columns.

The project can be used as a Python command-line tool, as a widget in Jupyter-style notebooks, or as npm components for React and Svelte. It also exposes an MCP interface so AI agents can query the schema, run SQL, and create charts.

Key features

  • 01Automatic clustering and labeling of embedding points
  • 02Kernel density estimation with density contours
  • 03Real-time search and nearest-neighbor lookup
  • 04WebGPU-based rendering for up to a few million points
  • 05Linked charts with cross-filtering across tabular columns
  • 06MCP access for AI agents to run SQL and create charts

Use cases

  • Exploring the structure and outliers of a text or image embedding dataset
  • Inspecting search results by finding neighbors of a known example
  • Adding an interactive embedding viewer to a notebook or web app
  • Letting an AI agent analyze and chart a tabular dataset

Quick start

  1. 1

    Install the Python package

    pip install embedding-atlas
  2. 2

    Launch the tool on your dataset

    embedding-atlas <your-dataset>

Things to know

  • Needs a WebGPU-capable browser for its smooth rendering performance.

Topics

#embedding#visualization#data-visualization#webgpu#clustering#machine-learning#notebook

FAQ

What is embedding-atlas?
Interactive visualization tool for exploring large embedding datasets along with their metadata, with cross-filtering and search. Embedding Atlas is a visualization tool for large sets of embeddings and the metadata attached to them. It lets you explore the overall structure of your data, see dense regions and outliers, and search for points similar to a query or an existing item.
Is embedding-atlas free to use? What license does it have?
embedding-atlas is open source under the MIT license. Check the license terms before using it in commercial products.
Is embedding-atlas still maintained?
Actively maintained. The last commit to embedding-atlas was on Oct 21, 2026.
What can I use embedding-atlas for?
Exploring the structure and outliers of a text or image embedding dataset Inspecting search results by finding neighbors of a known example Adding an interactive embedding viewer to a notebook or web app Letting an AI agent analyze and chart a tabular dataset
How do I install embedding-atlas?
Install the Python package: pip install embedding-atlas Launch the tool on your dataset: embedding-atlas <your-dataset>
Where can I get embedding-atlas?
The source code and installation instructions are on GitHub: https://github.com/apple/embedding-atlas

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects