Your data is hierarchical.
Your tools should be too.
HyperView is an open-source data curation co-pilot. Explore embeddings in hyperbolic space—where hierarchy has room to breathe—and let agents find the issues you'd otherwise miss.
$ pip install hyperview && hyperview demoWhy geometry matters
Hierarchical data gets crowded in Euclidean projections. The crowding problem means there isn't enough room to keep fine-grained structure separated. Rare modes collapse into dominant clusters—representation collapse.
Hyperbolic space has exponential volume growth. That matches hierarchy.
Correct interactions matter: Möbius pan/zoom and geodesic-aware selection.
Euclidean (flat)
V(r) ~ rd
Polynomial growth. Rare subgroups overlap.
see transformation →Hyperbolic (curved)
V(r) ~ er
Exponential growth. Hierarchy preserved.
see the difference →HyperView lets you toggle Euclidean ↔ hyperbolic (Poincaré disk) ↔ spherical—fast.
Projects
Open source. Code is MIT; packages on PyPI/npm. Model weights may carry upstream licenses.
Dual-panel curation UI: image grid + embedding map. Euclidean ↔ Poincaré ↔ spherical.
- →Agentic data cleanup
- →Multi-geometry views
- →HuggingFace integration
pip install hyperviewdemo Pure WebGL2 scatterplot for Euclidean + Poincaré disk with Möbius-correct interactions.
- →Möbius pan/zoom
- →Geodesic-aware lasso
- →20M points @ 60 FPS
npm i hyper-scatterdemo Non-Euclidean embedding encoders with simple API and torch-free ONNX runtime.
- →Hyperbolic encoders
- →Torch-free ONNX
- →Auto HF download
pip install hyper-modelsdemo