A CoCalc-AI field guide for working researchers

Installing Software in CoCalc-AI

Your project is a real Linux environment. When a command, library, or notebook dependency is missing, Codex can help you install it, verify it, and leave behind a reproducible trail that collaborators can inspect.

Illustrated workflow connecting a terminal, Codex assistant, package layers, verification, and a snapshot checkpoint

This guide follows one ordinary moment: you open a notebook, terminal, or project file, and something is missing. Maybe Python cannot import a package. Maybe a build script needs graphviz. Maybe a LaTeX document needs a system tool.

In CoCalc-AI, that is not a dead end. You have a Linux environment, a terminal, root access through sudo, and an agent that is very good at reading error messages and turning them into checked commands.

The unusual part is that this environment is collaborative. Terminal sessions, notebooks, files, chat, and Codex threads can all be shared live, so installing software becomes a project activity rather than a private ritual on one machine.

Know the layers

Software usually belongs in one of three places. Project files live beside your work. Language packages belong to Python, R, Node, TeX, or another toolchain. System packages live in the Ubuntu root filesystem.

Most install confusion comes from putting the right thing in the wrong layer.

Illustration of project files, language packages, and system packages as three layers in a CoCalc-AI environment

Start with the error

The best install request includes the exact failure. Do not begin with “install some stuff.” Begin with what broke.

This notebook cell says ModuleNotFoundError: graphviz. Please install what is needed, rerun a tiny verification, and tell me what changed.

Codex can inspect the terminal output, try the smallest fix, and verify the result before calling it done.

Illustration of Codex turning an error message into a narrow installation plan and verification checklist

Use apt for system tools

CoCalc-AI projects run in an Ubuntu environment. If you need a command-line tool or system library, the basic pattern is:

sudo apt-get update
sudo apt-get install -y graphviz
dot -V

The first line refreshes the package index. The second installs the package. The third proves the command exists.

Illustration of apt installing system packages into an Ubuntu root filesystem and ending with a verification check

Install language packages where the code runs

Python packages usually come from pip or conda. Node packages come from npm or pnpm. R packages come from R. TeX packages come from TeX Live or system packages.

Notebooks add one extra wrinkle: the terminal and the notebook kernel must be using the same environment. If an import still fails, restart the kernel or ask Codex to check which Python the notebook is actually using.

Illustration of language packages flowing to the matching notebook, terminal, and project environments

Use Codex as the installer, not the oracle

1 Show the failure

Paste the error or ask Codex to read the shared terminal history.

2 Install narrowly

Prefer the smallest package set that explains the error.

3 Verify directly

Run --version, import the module, or rerun the cell.

4 Write it down

Keep a short SETUP.md or setup script in the project.

Use snapshots as a safety net

Most installs are routine. Some are not. CoCalc-AI snapshots and backups give you a useful escape hatch: you can snapshot the whole container, including the root filesystem, before a risky change.

That means system-level experimentation does not have to feel permanent. If an install badly changes the environment, the root filesystem can be restored to an earlier version.

Simple habit: before a large system install, ask Codex to snapshot first and record what it changed afterward.

Illustration of taking a snapshot before a risky install and restoring the container root filesystem if needed

The real skill is knowing what changed

Installing software is not just making the error disappear. The good workflow is: diagnose, install, verify, document, and keep a way back.

CoCalc-AI works well here because the terminal, files, notebooks, snapshots, chat, and Codex all live in the same realtime collaborative project. The environment is not a black box; it is something a team can inspect and improve together.