Jupyter Hosting

How Jupyter works

Jupyter notebooks combine executable code cells, Markdown explanations, equations, visual output, and files in one browser-based document. A kernel runs code and returns results to the notebook interface, while JupyterLab adds tabs, a file browser, editors, consoles, and running-session controls. The catalogue starts Jupyter Server with a generated token and persists the /home/jovyan/work directory.

Access to a Jupyter Server is security-sensitive because an authenticated user can execute arbitrary code through kernels and other server features. The token must be protected like an administrator credential and should not be embedded in public links. The package is a single-user server, not JupyterHub with managed accounts or isolated environments.

Key Jupyter features

The selected quay.io/jupyter/base-notebook image includes JupyterLab, Notebook, NBClassic, and JupyterHub’s single-user component, but remains a minimal base. Official Docker Stacks documentation says it lacks the scipy-notebook scientific bundle and LaTeX PDF support. Users must install compatible packages within the environment or choose a different image outside this catalogue offer.

Only the work directory is declared persistent by the catalogue. Files saved elsewhere in the container may not survive replacement, and packages installed into non-persistent locations can disappear when the image changes. The plan also includes no GPU, TPU, accelerator subscription, shared notebook service, managed dataset catalogue, or Google Drive integration. CPU and memory-intensive notebooks remain limited by the selected hosting plan.

Jupyter vs Google Colab

Jupyter on this page is a self-hosted single-user base-notebook container with a persistent work directory and token authentication. Google Colab is a provider-managed notebook service with Google account sharing, hosted runtimes, and accelerator options subject to availability, quotas, plan terms, and session behaviour. Both use notebook workflows but differ sharply in infrastructure and collaboration model.

Colab may suit users who want a ready cloud runtime, Google Drive workflows, simple sharing, and optional managed accelerators without maintaining a server. The hosted Jupyter package is attractive when a stable custom domain and persistent work directory matter. It does not reproduce Colab’s account sharing, runtime catalogue, accelerators, preinstalled data-science environment, or provider-managed collaboration.

Who uses Jupyter

Learners write code beside explanations, developers test small scripts and APIs, analysts explore modest datasets, and instructors prepare demonstrations in reproducible notebooks. JupyterLab also supports text files and interactive sessions beyond .ipynb documents, making one workspace useful for experimentation that does not require a full multi-user platform.

Jupyter is not a safe public code paste, a sandbox for untrusted users, a GPU service, or an enterprise notebook hub. A notebook can read accessible files, consume plan resources, contact external services, and execute harmful code. Administrators should trust every person who receives the token and review dependencies before running notebooks from unknown sources.

Self-hosting Jupyter: requirements and cost

Jupyter resource use depends on active kernels, imported libraries, dataset size, in-memory objects, notebook outputs, files, concurrent computations, and external downloads. PostgreSQL and MariaDB are Not required in the catalogue template; notebooks and project files should be stored under the persistent work directory. A small learning notebook is light, while data processing can exhaust plan memory or CPU quickly.

On AvaHost, Jupyter uses Plan 1 at €2. The hosted Jupyter package includes one-click deployment, a custom domain with automated HTTPS, automatic application updates, and scheduled backups. Use the generated token as a sensitive credential and store project files under /home/jovyan/work. The package is based on jupyter/base-notebook; scientific stacks, LaTeX PDF export, GPU or TPU access, JupyterHub multi-user management, Google Drive integration, managed datasets, and durable files outside the mounted work directory are not included.