Building Module 6 · Your Agent In Depth: Desktop App & CLI

Your Agent Hub & Reference

Last reviewed

Beginner

What you'll learn

~10 min
  • Know that each of the four tools comes as a desktop app and as a terminal CLI that share one agent
  • Know where to look for npm installs, proxies and API keys if you ever need them
  • Use the Agent Tutor prompt to have your own agent coach you through lessons
  • Open the in-depth lesson for the one tool you installed in Lesson 1.2
đź’ˇAlready running an agent from Lesson 1.2?

Then you’re set up. This module takes the tool you have deeper: its plans and models, its modes and what each allows, switching between its desktop app and its CLI, and troubleshooting. Arriving here from a track start with nothing installed yet? Pick a tool below and install it from its lesson; Lesson 1.2 is the short, step-by-step version.

ℹUsing a cloud sandbox?

Each tool lesson includes a “Cloud Equivalent” callout with the browser-based alternative, and if you’re in Codespaces you can install the CLIs directly in your codespace. The domain track lessons work in Codespaces too, so you don’t need a local install to continue. If you only use Claude Code Web, you can go straight to Module 7: Git.

Two ways in: desktop app or terminal

Each of the four makers ships its agent in two local forms: a desktop app you click through, and a command-line tool (CLI) you type into. Both open a folder on your computer, read and edit the files in it, run commands, and plan with you before they build. They also read the same project files, so a CLAUDE.md or AGENTS.md brief, a skill, or an MCP server you set up in one works in the other. What you learn in Modules 7 to 20 (planning, reviewing, checkpoints, project memory) carries across both.

MakerDesktop appTerminal CLIFree way in
AnthropicClaude desktop app, Code tabclaudeNone: Pro or a higher paid plan
GoogleAntigravity appagyFree tier, under weekly limits
OpenAIChatGPT desktop app, CodexcodexFree and Go: GPT-6 Luna in the desktop app
GitHubGitHub Copilot appcopilotCopilot Free

There is no wrong first door. A desktop app needs no terminal at all, and every CLI also installs natively on Windows (in PowerShell), macOS and Linux. When you want more speed, a CLI on Linux, WSL2 or macOS is where builds run fastest and where tmux lets you keep several agents working side by side. Step up when your current setup feels like the slow part, not before. Module 4’s remote environments lesson explains the course author’s own Linux and tmux setup, if you want to judge whether that step is worth it for you.

ℹPre-flight check

You need a terminal that opens (Terminal on macOS, Windows Terminal or PowerShell on Windows, or your Linux shell) and Git (git --version prints a version number). Using only a desktop app? Skip the terminal steps. If Git is missing, install your tool first and ask it in plain English to install Git for you. A code editor such as VS Code is optional.

Anything else your project needs (a language runtime, a package, a starter file) is something you ask the agent for in plain English. You don’t need Node.js or API keys to follow this module.

Advanced: npm installs, corporate proxies & API keys

You only need this section if you choose an npm install, work behind a corporate proxy or SSL inspection, or sign in with an API key instead of an account.

Node.js and npm

Node.js Node.js A platform that runs JavaScript outside the browser. Learn more → , a platform that runs JavaScript outside the browser, is required only for the npm route of Copilot CLI (Node 22 or newer). npm is a secondary install path for Claude Code and Codex CLI, and Antigravity’s native binary needs no Node.js at all. Check what you have:

Terminal window
node --version
npm --version

v22 or higher is fine (Node 24 or 26 LTS too). To install it on macOS, Linux, or WSL without admin rights, use nvm (Node Version Manager):

Terminal window
curl -o- https://raw.githubusercontent.com/nvm-sh/nvm/v0.41.0/install.sh | bash

Then close and reopen your terminal (the script adds itself to ~/.bashrc or ~/.zshrc, which only new terminals read) and run:

Terminal window
nvm install --lts
node --version

If you can’t use nvm, brew install node works on macOS, and nodejs.org has LTS installers for every OS. Ubuntu’s own apt package is often older than v22; use the NodeSource repository instead.

npm install -g tool installs a tool permanently; npx tool downloads and runs it once. If npm install -g fails with EACCES (permission denied), Node was installed system-wide: switch to nvm rather than reaching for sudo.

API keys

An API Key API Key A secret string that authenticates your account with an AI service. Learn more → is a secret string that signs a tool in to an AI service and bills usage to your account. Every tool in this module also supports account sign-in, which most learners should use. If you use a key:

  • Never share it, paste it into a public chat, or put it in a file that gets committed to git (add .env to your .gitignore).
  • If you accidentally expose a key, revoke it immediately from the provider’s dashboard.
  • Most tools store your key for you. To set one by hand for the current session:
Terminal window
export ANTHROPIC_API_KEY="sk-ant-your-key-here"

To keep it, add that line to ~/.bashrc (or ~/.zshrc on macOS). On a shared machine, check that file’s permissions with ls -la ~/.bashrc: they should read -rw-------; if not, run chmod 600 ~/.bashrc.

🧬In Your Field: Biotechclick to expand

If you’re on a shared server or lab workstation (e.g., a lab’s analysis node), multiple people may have accounts on the same machine. Always check file permissions on your shell config, and consider using a .env file in your home directory with chmod 600 instead of putting keys directly in .bashrc. This way, you can revoke one project’s keys without affecting your shell.

Signing in over SSH

Some tools want to open a web browser to sign in. Over SSH there’s no browser, so most tools show a URL and a code (the device flow): open the URL on your own machine, enter the code, and the remote tool is signed in. An API key set as an environment variable before you start the tool skips the browser entirely. Each tool lesson covers its own SSH sign-in.

Corporate proxies and certificates

On a university network, corporate VPN, or government network, npm installs can fail with certificate errors (SELF_SIGNED_CERT_IN_CHAIN, which means the network inspects SSL traffic) or time out. Point npm at your organization’s certificate bundle, or at its proxy (ask your IT team for the address):

Terminal window
npm config set cafile /path/to/your/corporate-ca.pem
npm config set proxy http://proxy.example.com:8080
npm config set https-proxy http://proxy.example.com:8080

Or use GitHub Codespaces to sidestep network restrictions entirely.

🏛️In Your Field: Government Devclick to expand

Highly secure corporate or government networks often block outbound connections to registries like npmjs.org. If your organization uses an internal package mirror or Artifactory instance, you’ll need to configure npm to use it: npm config set registry https://your-agency-mirror.example.gov/npm/. Check with your IT security team; they may also require you to use pre-approved tool versions from an internal catalog rather than installing directly from the public registry.

đź”§

When Things Go Wrong

Use the Symptom → Evidence → Request pattern: describe what you see, paste the error, then ask for a fix.

Symptom
nvm: command not found (after running the install script)
Evidence
bash: nvm: command not found
What to ask the AI
"I installed nvm using the curl script but my terminal says 'nvm: command not found.' I'm using bash on WSL/macOS/Linux. How do I fix this?"
Symptom
npm install -g fails with EACCES permission error
Evidence
npm ERR! Error: EACCES: permission denied, mkdir '/usr/local/lib/node_modules'
What to ask the AI
"I get a permission denied error when running npm install -g. I installed Node.js with apt instead of nvm. What's the safest way to fix this without using sudo for every install?"
Symptom
npm install fails or times out on a restricted network
Evidence
npm ERR! code SELF_SIGNED_CERT_IN_CHAIN, or a connection timeout to registry.npmjs.org
What to ask the AI
"npm installs fail on my university/corporate network with a certificate error or a timeout. How do I configure npm to use our proxy or certificate bundle?"

After you install: a tutor in your own agent

Once your agent is installed, a separate session in it can coach you through any lesson: you copy one prompt, and that coaching session reads the tutor page and the lesson, then walks you through it one step at a time while you run commands or builder prompts yourself in a second tab or window. The card with that prompt sits at the bottom of Lesson 1.2 and each tool lesson in this module, and folded on lessons from Module 2 on. Open the practice folder the lesson names (or create the lesson folder shown on the card) and start your coaching session there.

Before you paste the prompt, switch that coaching session to plan mode as its own step (Shift+Tab, /plan, or your desktop app’s mode picker; the tool modes table has the details). If your agent asks permission to fetch learn.intellimetrics.net (including plain codex asking to fetch outside its sandbox), approve it. When you pass a check and explain what it proved, the coach gives you one line for PROGRESS.md, and you add it yourself.

Pick your tool

Pick the one tool you installed in Lesson 1.2; completing any one of the four finishes Module 6 and steps straight to Module 7.


Key Takeaways

  • Two ways in, one agent: each maker ships a desktop app and a terminal CLI that share the same engine and project files
  • Start where you’re comfortable: a desktop app, or a CLI in your own terminal (PowerShell included), is a complete first path; step up to Linux, WSL2 or macOS when you want speed
  • One tool is enough: pick the tool you installed in Lesson 1.2 and go deep
  • npm, proxies and API keys are optional: they live in the Advanced section for the cases that need them
Search lessons