# getkeops/keops

KErnel OPerationS, on CPUs and GPUs, with autodiff and without memory overflows

Repository: https://github.com/getkeops/keops
Canonical: https://ross.abutalabs.com/products/keops
Homepage: https://www.kernel-operations.io
Language: Python
License: MIT
License Family: permissive
Last push: 2026-07-17T15:05:52+00:00

## Health v2 (maintenance only)
Score: 65/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 93, release rhythm 8, longevity 100
- inputs: {"age_days": 2694, "days_push": 47, "days_rel": 502, "gap_med": null, "n_releases_24m": 1}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1189, forks 78 (observed 2026-08-28T04:03:55.412670+00:00)

## What it is
KeOps (pykeops) is a Python library for computing kernel reductions over large arrays on CPUs and GPUs using efficient C++/CUDA routines with automatic differentiation. It avoids materializing full kernel or distance matrices, enabling memory-efficient KNN queries, N-body interactions, and point cloud convolutions.

## Use cases
- compute kernel matrix-vector products that don't fit in GPU memory
- fast K-nearest neighbors search on large point clouds
- N-body interaction simulations with gradients
- point cloud convolutions for geometric deep learning
- autodiff through kernel operations in PyTorch
- speed up Gaussian kernel computations by 10-100x over PyTorch

## When to choose
- you need kernel or distance computations on millions of points without memory overflows
- you want autodiff-capable geometric operations in NumPy, PyTorch, Matlab, or R
- geometric deep learning or kernel methods are memory-bound in your pipeline

## When to avoid
- you need simple dense linear algebra that standard BLAS/PyTorch handles fine
- your project requires pure Python without C++/CUDA compilation
- you work on small datasets where the overhead isn't worth it

## Facets
- artifact type: library
- maturity: active
- function: math, machine-learning, gpu-computing, benchmarking
- domain: machine-learning, gpu-computing, mathematics
- platform: python, cross-platform
- tags: kernel-operations, autodiff, cuda, pytorch, numpy, nearest-neighbors, point-clouds, symbolic-tensors, algorithms, gpu

## Member repositories
- getkeops/keops (main) score 65

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:03:55.412670+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T06:23:09.298735+00:00, confidence not recorded.
  - readme: https://github.com/getkeops/keops (fetched 2026-08-28T04:03:55.412670+00:00, sha 7014761119a7)
  - homepage: https://www.kernel-operations.io (fetched 2026-08-29T12:30:45.447814+00:00, sha 44136fa355b3)
- Data as of 2026-08-30T08:39:29.467469+00:00.