Open core

kernel-torch
Premium Optimiser

An open-core probabilistic machine-learning framework. Compose a loss landscape out of declarative terms, hand it to one optimiser, and let it handle the rest — built for AI/ML, deep learning and Bayesian inference developers.

$ pip install kernel-torch

One optimiser, composable objectives

Stochastic Control

Describe your loss landscape — priors, penalties and all. We find the optimal allocation. Think of kernel-torch as the autograd to your autograd.

Natively Bayesian

Variational Posteriors, Langevin Monte Carlo sampling, Multi-latent Gaussian Processes. Nothing is point-wise. Quantifiable uncertainty is a feature, not a bug.

Tensor Logic (Pytorch/Candle)

Ordinary tensors, ordinary autograd, ordinary CUDA. It slots into training code you already have — no runtime to adopt, no graph to rewrite. We make your life easier, not harder.

Open source

The framework

The client, the loss-landscape algebra, and the optimiser are open and stay that way. Run it entirely on your own hardware and never talk to us.

View on GitHub →

Premium

The engine

Point the same code at our infrastructure when a fit outgrows your box: distributed bridge training, hosted inference, and the trained-kernel catalogue — behind one API key.

See plans →

Start local. Scale when it hurts.

Install the framework today; add an API key the day a fit stops fitting.

Create account