Kubeflow Notebooks
kubeflow.org
- Free plan
- Web
- Self-hosted
What Kubeflow Notebooks is
Kubeflow Notebooks provides hosted, interactive environments for AI, machine learning, and data work inside Kubernetes clusters. Users can create notebook containers in the cluster rather than running them on local workstations. Native interfaces include JupyterLab, RStudio, and Visual Studio Code through code-server; Python, R, and Jupyter Notebook files are supported, and other web-based IDEs should also work. Notebook servers can request CPU, RAM, workspace and data volumes, and optional GPUs. Administrators can publish standard images with required packages, while Kubeflow RBAC manages organizational sharing. Packages installed after startup last only for the pod unless stored in a PVC-backed directory or custom image. Custom images must expose port 8888, run as jovyan with UID 1000, and use /home/jovyan. The project is open source, free, and deployable standalone or within a Kubeflow distribution.
Who it suits
It suits AI practitioners and teams running shared notebook environments on Kubernetes. Platform administrators can use RBAC, standard images, configurable resources, storage, and GPU requests.
Good to know, and what to watch
- Runs notebook containers directly in Kubernetes
- Supports JupyterLab, RStudio, and code-server
- Allows CPU, RAM, volume, and GPU requests
- Administrators can provide standard images
- Open source and free
- Requires Kubernetes deployment
- Packages may disappear when pods restart
- Custom images must meet specific requirements
- Community support is best effort
The verdict
Kubeflow Notebooks centralizes configurable development environments for teams operating Kubernetes infrastructure. Persistent packages and compatible images require deliberate setup.
What the maker publishes
- Access control
- Notebook access control is managed by Kubeflow RBAC to support sharing across an organization. source
- Access control
- Notebook access control is managed by Kubeflow RBAC, which the docs say enables notebook sharing across an organization. source
- Container images
- The project documents notebook images for code-server with Python and for JupyterLab with PyTorch, TensorFlow, CUDA, and common packages. source
- Container images
- Official example images include JupyterLab with PyTorch or TensorFlow, RStudio with Tidyverse, and code-server with Conda Python. source
- Custom image requirements
- A compatible custom image must expose HTTP on port 8888, run as user jovyan with home directory /home/jovyan and UID 1000, and start with an empty PVC mounted there. source
- Deployment
- Kubeflow Notebooks can be deployed as a standalone subproject or as part of a Kubeflow distribution. source
- Image requirements
- Custom notebook images must expose an HTTP interface on port 8888, run as user jovyan with UID 1000, and start with an empty PVC mounted at /home/jovyan. source
- In-cluster environments
- Users can create notebook containers directly in the cluster instead of running them locally on their workstations. source
- Intended users
- Kubeflow describes its broader platform as serving AI practitioners, platform administrators, and decision-makers with tools for data, AI/ML, and HPC workloads. source
- Maturity
- The installation guide lists Kubeflow Notebooks among graduated projects, which it describes as stable and ready for general availability. source
- Notebook interfaces
- It natively supports JupyterLab, RStudio, and Visual Studio Code through code-server. source
- Organization images
- Administrators can provide standard notebook images with required packages pre-installed. source
- Organization images
- Administrators can provide standard notebook images with required packages pre-installed, and users can also select a custom Docker image. source
- Package persistence
- Packages installed after a notebook starts last only for the lifetime of its pod unless installed in a PVC-backed directory; custom images or PVC-backed installation preserve them through pod restarts. source
- Package persistence
- Packages installed after a notebook starts last only for the lifetime of its pod unless installed in a PVC-backed directory; custom images are another way to preserve packages across pod restarts. source
- Project governance
- Kubeflow is part of the Cloud Native Computing Foundation and follows its governance structure. source
- Purpose
- Kubeflow Notebooks runs interactive development environments for AI, ML, and data workloads on Kubernetes. source
- Purpose
- Kubeflow Notebooks runs interactive development environments for AI, ML, and data workloads on Kubernetes. source
- Related beta project
- Kubeflow Workspaces is described as the next generation of Notebooks (Notebooks v2) and is beta software intended for staging and development clusters. source
- Resource configuration
- Users can set CPU and RAM requests, mount workspace and data volumes, and optionally request GPUs when creating a notebook server. source
- Resources and storage
- When creating a notebook server, users can specify CPU, RAM, a workspace volume, optional data volumes, and optional GPU requests. source
- Runs in cluster
- Notebook containers run directly in a Kubernetes cluster rather than on users’ workstations. source
- Secrets configuration
- Notebook configurations can match PodDefault resources in a profile namespace, including a documented example for mounting a GCP credential secret. source
- Support
- The Kubeflow community offers best-effort support for stable and beta applications, and the project lists commercial support providers including Canonical and Red Hat. source
Checked against the maker's pages on 9 Oct 2026.