workstation-setup

Ansible playbooks for provisioning a personal Ubuntu workstation for software development, AI/ML work, and general engineering use.

Designed to be repeatable — safe to run multiple times on the same machine.

Target environment

• Ubuntu 26.04
• Local workstation
• NVIDIA GPU (optional, RTX 5070 primary target)

Quick start

Full setup with CUDA:

./bootstrap.sh

Setup without CUDA:

./bootstrap-without-cuda.sh

Both scripts install Ansible Galaxy requirements and run the appropriate playbook under sudo.

Repository layout

bootstrap.sh               # Full CUDA setup entry point
bootstrap-without-cuda.sh  # Non-CUDA setup entry point

playbooks/
  site-cuda.yml            # Full workstation + NVIDIA/CUDA
  site.yml                 # Full workstation, no GPU assumption
  nvidia-rtx5070-ubuntu2604.yml  # Standalone GPU driver installer

roles/
  base/           # Core apt packages, ChatGPT desktop
  user/           # Dev dirs, git config, shell (Starship, aliases, tealdeer)
  docker/         # Docker CE, Compose, Buildx; adds user to docker group
  devtools/       # lazygit, VS Code, Sublime Text, Cursor IDE, tmux, httpie
  extras/         # Spotify
  jvm/            # SDKMAN, Java 21, Gradle, Groovy
  nvidia_cuda/    # NVIDIA driver, CUDA Toolkit, environment, validation
  python_ai/      # pyenv, uv, Python 3.11, AI venv (PyTorch, scikit-learn, etc.)
  security/       # Bitwarden desktop
  node_web/       # nvm, Node.js LTS, pnpm

Key variables

Defined in playbooks/site-cuda.yml and playbooks/site.yml:

Variable	Default	Description
dev_user	sampo	Target user account
python_version	3.11.13	pyenv-managed Python version
java_version	21.0.11-tem	SDKMAN Java version
gradle_version	9.5.0	SDKMAN Gradle version
groovy_version	4.0.31	SDKMAN Groovy version
cursor_path	/opt/cursor.AppImage	Cursor IDE install path
nvidia_cuda_manage_driver	true in cuda playbook	Install the NVIDIA driver
nvidia_cuda_install_toolkit	true in cuda playbook	Install CUDA Toolkit
nvidia_cuda_validate	true	Run nvidia-smi check
nvidia_cuda_driver_package	nvidia-driver-595-open	Driver package name

What each role does

base — installs core CLI tools (ripgrep, fd, bat, fzf, jq, tree, ncdu, tealdeer, vim, etc.), zram-config, gedit, gnome-tweaks, and the ChatGPT desktop snap.

user — creates ~/dev, ~/bin, ~/tmp, ~/.bashrc.d, configures global git settings, sets up Starship prompt, and installs tealdeer with auto-update.

docker — installs Docker CE from the official Docker repo, enables the service, and adds dev_user to the docker group. Requires logout/login to take effect.

devtools — installs tmux, httpie, lazygit, VS Code (snap, classic), VS Code extensions (GitHub Copilot, Claude Dev), Sublime Text, and Cursor IDE (AppImage downloaded from the Cursor API).

extras — installs Spotify via snap.

jvm — installs SDKMAN for the user and uses it to install Java, Gradle, and Groovy at the pinned versions.

nvidia_cuda — conditionally installs the NVIDIA driver and CUDA Toolkit, writes CUDA env vars to /etc/profile.d/cuda.sh, and validates the install with nvidia-smi and nvcc.

python_ai — installs pyenv and uv for the user, installs Python 3.11, creates an AI sandbox venv at ~/dev/ai-sandbox, and installs PyTorch (CUDA 12.8 wheel), ultralytics, scikit-learn, opencv, numpy, pandas, matplotlib, and JupyterLab.

security — installs Bitwarden desktop via snap.

node_web — installs nvm for the user, installs Node.js LTS, and installs pnpm globally.

Running individual parts

Override variables at the command line:

# Install only the GPU driver, skip everything else
sudo ansible-playbook playbooks/nvidia-rtx5070-ubuntu2604.yml

# Run full playbook but skip NVIDIA driver install
sudo ansible-playbook playbooks/site-cuda.yml -e nvidia_cuda_manage_driver=false

# Run full playbook but skip CUDA Toolkit
sudo ansible-playbook playbooks/site-cuda.yml -e nvidia_cuda_install_toolkit=false

Validation

# GPU
nvidia-smi
nvcc --version

# Python + GPU
python - <<'PY'
import torch
print("torch:", torch.__version__)
print("cuda:", torch.cuda.is_available())
print("gpu:", torch.cuda.get_device_name(0) if torch.cuda.is_available() else None)
PY

# Docker
docker run --rm hello-world

# Node
node --version && npm --version

# Java
java --version