Sandbox

Base Image

What every Tenki Sandbox session ships with out of the box, and how to add your own tools.

Every Tenki Sandbox session starts from the same prepared base image, so common runtimes and tools are already present the moment a session reaches READY. The essentials need no install-and-warm-up step. This page lists what ships by default and how to add anything else.

The base is a standard Ubuntu 24.04 LTS microVM. You log in as the tenki user (uid 1000) with passwordless sudo, so you can apt install, pip install, or npm install anything not listed here.

Versions change over time

The versions below reflect the current base image and are refreshed as tools release updates. To see exactly what a live session has, run the version command for the tool (for example python3 --version or node --version) inside the session.

Language runtimes

RuntimeVersionNotes
Python3.12Ubuntu 24.04 system Python, with pip
Node.js24.14.0Installed via nvm; npm included
Bun1.3.14Fast JavaScript/TypeScript runtime and bundler

Package managers & language tooling

ToolVersionPurpose
piplatestPython package installer
uv0.11.28Fast Python installer and resolver
npmbundledNode package manager (with Node.js)
pnpm10.20.0Fast, disk-efficient Node package manager
TypeScript (tsc)5.9.3TypeScript compiler
ts-node10.9.2Run TypeScript directly
typescript-language-server5.3.0TypeScript/JavaScript LSP

Coding agents

Tenki Sandbox bakes in the major terminal coding agents so you can drive them directly inside a session:

AgentCommandVersion
Claude Codeclaude2.1.209
OpenAI Codexcodex0.144.4
opencodeopencode1.17.20

Python data & ML libraries

Preinstalled into the system Python so data and analysis workloads run without a setup step:

CategoryPackageVersion
Data corenumpy2.4.1
Data corepandas2.3.3
Data corematplotlib3.10.8
Data corescipy1.17.0
Data coreseaborn0.13.2
Data corepillow12.1.0
Data corerequests2.32.5
Data corebeautifulsoup44.14.3
MLscikit-learn1.8.0
MLopencv-python-headless4.13.0.90

Heavier deep-learning frameworks (PyTorch, TensorFlow, Transformers, and similar) are not preinstalled. They tend to be large and GPU-oriented, so it's best to pin them per workload. Install them per session or bake them into a template.

Common CLI tools

Available on PATH in every session:

  • Version control & GitHub: git, gh
  • HTTP & transfer: curl, wget, rsync
  • Search & data: ripgrep (rg), jq, sqlite3
  • Archives: tar, zip/unzip, zstd, bzip2, xz
  • Build: build-essential (gcc/make), pkg-config
  • SSH: openssh-client, openssh-server
  • Inspect: file, less, lsof

Checking what's installed

Because you have sudo, you can inspect anything from inside a session:

python3 --version
node --version
claude --version
python3 -c "import numpy, pandas, sklearn; print(numpy.__version__, pandas.__version__)"
dpkg -l | grep -i ripgrep

Adding your own tools

For a one-off session, just install what you need:

pip install polars
sudo apt-get update && sudo apt-get install -y ffmpeg
npm install -g some-cli

If every session should start with the same extra tools, define them once in a template instead of installing on each session — the template build captures the prepared environment as a snapshot so new sessions start ready.

Need a package added to the base?

If you think a tool or library is broadly useful enough to belong in the default base image, let us know. We keep the base intentionally lean, but the included package set evolves based on what customers reach for most.