Templates¶

Templates are YAML files that define what software is installed when you launch a workspace.

Using templates¶

# List available templates
prism templates

# Details on a specific template
prism templates info python-ml

# Launch a workspace from a template
prism workspace launch python-ml my-project
prism workspace launch r-research stats-project --size L

Built-in templates¶

Template	What's included
`python-ml`	Python, PyTorch, TensorFlow, Jupyter
`r-research`	R, RStudio Server, Bioconductor
`genomics`	GATK, BWA, samtools, STAR
`bioinformatics`	Conda, Snakemake, BioPython
`deep-learning`	CUDA, cuDNN, GPU-ready PyTorch
`data-science`	Pandas, scikit-learn, DuckDB
`hpc-base`	MPI, OpenMP, GCC, CMake

Template YAML format¶

Templates live in the templates/ directory and are written in YAML.

Minimal example¶

name: my-template
description: Python environment with data science tools
base: ubuntu-22.04-server-lts
architecture: x86_64

packages:
  - python3
  - python3-pip
  - jupyter

build_steps:
  - name: Install Python packages
    script: |
      pip3 install numpy pandas scikit-learn matplotlib

validation:
  - name: Check Python
    script: python3 --version
  - name: Check Jupyter
    script: jupyter --version

Full schema¶

Field	Required	Description
`name`	Yes	Unique template name
`description`	Yes	Human-readable description
`base`	Yes	Base OS image (e.g. `ubuntu-22.04-server-lts`)
`architecture`	No	`x86_64` (default) or `arm64`
`inherits`	No	Parent template name (see Inheritance below)
`package_manager`	No	`apt`, `conda`, or `dnf`
`packages`	No	List of packages to install
`users`	No	Additional OS users to create
`services`	No	Services to start (e.g. `jupyter`, `rstudio`)
`ports`	No	Ports to open in the security group
`build_steps`	No	Ordered list of setup scripts
`validation`	No	Tests run after build to verify the template works

Build steps¶

Each build step runs a shell script:

build_steps:
  - name: Install R packages
    script: |
      Rscript -e "install.packages(c('tidyverse', 'ggplot2'), repos='https://cran.r-project.org')"
    timeout: 1800    # seconds; default 600

Validation¶

Validation runs after build. A non-zero exit code marks the template as broken:

validation:
  - name: R is installed
    script: R --version
  - name: tidyverse loads
    script: Rscript -e "library(tidyverse)"

Template inheritance¶

Templates can inherit from a parent. The child adds to (not replaces) the parent's packages, users, services, and ports. The child's package_manager replaces the parent's.

name: python-ml-gpu
description: python-ml with GPU support
inherits: python-ml

packages:
  - cuda-toolkit-12
  - cudnn9

build_steps:
  - name: Install GPU PyTorch
    script: pip3 install torch --index-url https://download.pytorch.org/whl/cu121

Merging rules: - packages, users, services: append (child adds to parent) - ports: deduplicate - package_manager: override (child replaces parent) - build_steps: append (parent steps run first)

Creating a custom template¶

Create a YAML file in templates/:

cp templates/python-ml.yaml templates/my-template.yaml

Edit the file with your changes.
Validate the template:
```
prism templates validate my-template
```

Launch a test workspace:

prism workspace launch my-template test-build

Tips¶

Start from an existing template rather than from scratch — inheritance saves a lot of work.
Only include packages you actually need; smaller templates launch faster.
Add validation steps for the tools your users will rely on most.
Set timeout on long build steps (R package installs, compiling from source).