OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
AI Infrastructure & Serving

Kubetorch

Kubetorch is an open-source Python client and Kubernetes deployment stack for running machine-learning workloads on remote cluster compute. Developers can attach a local Python function to a compute request and call it remotely, with logs, exceptions and hardware faults propagated back to the calling environment. Its repository contains the SDK, Helm chart, controller and data-store sources, and workload base images.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Kubetorch — OpenViST project profile in AI Infrastructure & Serving.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Consider Kubetorch when ML code needs remote Kubernetes compute but developers want to define and call workloads from Python. Its documented pattern covers function dispatch; cluster installation is a separate Helm deployment step.

Who it is for

  • Machine-learning developers who want to run Python workloads on Kubernetes from an IDE or notebook.
  • Teams building distributed-training, reinforcement-learning or inference workflows on cluster compute.
  • Platform teams deploying a self-hosted ML workload runner with Kubernetes and Helm.

Documented uses

  • Send a locally defined Python function to remote Kubernetes compute.
  • Use cluster resources from notebooks, IDEs, CI pipelines or production Python code.
  • Receive remote logs, exceptions and hardware faults in the development workflow.
  • Deploy Kubetorch’s client and Kubernetes-side components as a self-hosted stack.
Before you choose

Requirements & limits

The README documents a pip-installable client and Helm-based self-hosted deployment. It also describes a fully managed serverless platform and directs interested users to contact the project team.

  • A Kubernetes cluster and Helm are needed for the documented self-hosted deployment.
  • The Python client and cluster-side stack are separate installation components.
  • The README provides a function-dispatch example, but cluster-specific configuration is left to the installation documentation.
  • The managed serverless offering is presented separately from the open-source deployment.
  • The repository states an Apache 2.0 license.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Install the Python client and Kubernetes components

Set up Kubetorch’s Python client and install its Kubernetes deployment with Helm.

Before you start

  • A Python environment with pip.
  • A Kubernetes cluster and Helm.
  • Access to the Kubetorch chart OCI registry.

Steps

  1. Install the client with the command below.

    Documented command
    pip install "kubetorch[client]"

  2. Install the chart from the OCI registry with helm upgrade --install kubetorch oci://ghcr.io/run-house/charts/kubetorch --version 0.5.0 -n kubetorch --create-namespace. Alternatively, pull and untar the chart, then install it from the local ./kubetorch directory.

Expected outcome

The Python client is installed and the Kubetorch Kubernetes components are deployed through Helm.

Guide limitations

  • This covers installation only; it does not verify a successful deployment or provide cluster-specific configuration.
  • The README points to the installation guide for detailed setup instructions.
Documentation 1 ↗
Source-based practical guide

Call a Python function on remote Kubernetes compute

Use the README’s hello-world pattern to send a local Python function to remote compute and call it.

Before you start

  • The Kubetorch Python client and Kubernetes deployment.
  • A Python environment able to import kubetorch and connect to the configured cluster.

Steps

  1. Import kubetorch as kt and define a Python function.

  2. Create a compute request with kt.Compute(cpus=".1").

  3. Attach the function using kt.fn(hello_world).to(compute).

  4. Call the returned function; the README’s example prints the returned string.

Expected outcome

The example function runs remotely on the Kubernetes cluster and returns a value to the calling Python code.

Guide limitations

  • The example assumes a configured Kubernetes deployment; the README does not give cluster configuration details in this example.
  • The recipe reflects the documented example and is not an execution or compatibility test.
Documentation 1 ↗
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
3 October 2026
Original website title
run-house/kubetorch · Official homepage title · retained in its source language
How we review sourcesSuggest a correction