4.9k stars · MIT · v0.24.0 (2026-08-20)
Apple's interactive embedding map with automatic cluster labelling — recommended as a design study for Grasppy's overview screen, not as a dependency.
▶Repo detailsthe review · specs · pros & cons · install
What it is
Apple's interactive visualisation for large embedding sets — a density map of up to millions of points with automatic cluster labelling, search, and cross-filtering against the metadata columns. It ships as a Python package you point at a dataframe, and as embeddable web components.
Why it matters
Treat this as a reference before you design the overview screen, not as a dependency you adopt. The hardest unsolved question in Grasppy's interface is what a user sees in the first second — a wall of nodes reads as noise, and the difference between a toy and a tool is whether that first view is legible without instruction. Apple's machine-learning group spent real design effort on precisely that: labels that fade in and out as you zoom, density instead of scattered dots, filtering that keeps you oriented rather than lost. Run it once over the embeddings of a conversation you already have and you will learn more about your own overview screen in an hour than in a week of sketching. It is also a fair candidate against reagraph from yesterday if you decide to build the overview on someone else's canvas.
- MIT, and from Apple's ML group — unusually considered interaction design for an open-source visualisation
- Point it at a pandas dataframe and it renders. The evaluation genuinely costs one afternoon
- Handles very large point counts smoothly, which is the exact failure mode of naive canvases
- It draws a point cloud, not a hierarchy. Grasppy's drill-in is a tree, and this does not give you one
- Adopting it wholesale means embedding a fairly opinionated tool rather than building the view you want
- Comparatively young — 161 commits, and the component API can still move under you
Apple's machine-learning group spent real design effort on precisely that: labels that fade in and out as you zoom, density instead of scattered dots, filtering that keeps you oriented rather than lost.
source venv/bin/activate pip install embedding-atlas