Lab Setup: Accounts and Your Coding Agent

How long does it take to go from "I've never opened a terminal" to an AI agent that writes files for you? About 15 to 30 minutes, if you skip the detours. By the end of this lesson you will have three free platform accounts (GitHub for your code, Vercel for hosting, Supabase for the database), one working coding agent, and proof that it works: an empty dayboard folder in which the agent has written its first file, HELLO.md, reporting the versions of Node.js, git and itself. DayBoard is the morning dashboard you'll build across the course, with local weather next to your task list. Today you won't write any app code. You'll set up the workbench and make sure every tool answers when you call it. You'll pick one of three setups: Claude Code on your laptop, OpenAI's Codex CLI on your laptop, or xAI's Grok Bot installing and driving one of those two on its own cloud computer. Each setup has a different security trade-off, so you'll also see who approves what the agent does, and which secrets must never go where.

Pick one of three setups for running your coding agent

Every setup ends with the same thing: a coding agent (Claude Code or Codex) running in a terminal, signed in with your own paid plan. The difference is whose computer that terminal is on.

If you're unsure, pick A or B, whichever matches a plan you already pay for. Grok Bot is for people who specifically want the agent off their laptop and want to hand off work and come back later (xAI).

The full comparison comes later. Lesson 01-05 covers harnesses and models, lesson 01-06 covers benchmarks, and lesson 01-07 helps you settle on a setup. For now, pick quickly.

Create your three platform accounts: GitHub, Vercel, Supabase

Do this part in a web browser, whichever setup you chose. Order matters: create GitHub first, because Vercel and Supabase both let you sign up with your GitHub account, which leaves you one less password to manage (Vercel sign-up, Supabase sign-up).

GitHub: where your code lives (turn on 2FA now)

1. Go to the GitHub sign-up page linked from GitHub's account guide and follow the prompts.
2. Verify your email. GitHub won't let you do basic tasks such as creating a repository until you do (GitHub Docs).
3. Turn on two-factor authentication (2FA). Open your profile picture → Settings → Password and authentication. GitHub recommends an authenticator (TOTP) app over SMS, because SMS can be intercepted (GitHub Docs: 2FA).
4. Save the recovery codes somewhere safe, such as a password manager. They're how you get back in if you lose your phone, and GitHub makes you confirm you've saved them before it turns 2FA on (GitHub Docs: 2FA).

Why 2FA now? Later in the course you'll connect GitHub to Vercel and give an agent scoped access to a repo. Your GitHub account becomes the key to everything else.

Vercel Hobby and Supabase Free: sign up with GitHub

Vercel (hosting). Open vercel.com/signup and choose Continue with GitHub. The Hobby plan is free and meant for personal projects (Vercel Hobby docs). Note one limit now: Hobby is restricted to non-commercial, personal use (Vercel Hobby docs). DayBoard as a learning project fits. If you later charge money for an app, that's a Pro-plan conversation (lesson 07-04).

Supabase (database and logins). Open supabase.com/dashboard/sign-up and choose Continue with GitHub. You don't need to create a project yet. You'll do that in Module 05. Two Free-plan facts to know early (checked 2026-10-20): you get 2 active projects, and free projects are paused after 1 week of inactivity (Supabase pricing). If DayBoard's database seems to vanish after a quiet week, it's probably paused, not deleted.

Laptop track, step 1: terminal, Node.js and git

Grok Bot track: skip to the Grok Bot section below.

Open a terminal on macOS, Windows or Linux

A terminal is a text window where you type commands. Anthropic's terminal guide for new users describes these steps:

On Windows, make sure you're in PowerShell, not CMD. CMD's prompt has no PS, and the two need different commands (Anthropic terminal guide).

Install Node.js LTS and git

Node.js runs the JavaScript tooling your agent will use to build and run DayBoard, and the Vercel CLI requires Node.js 18 or later (Vercel getting started). Install the LTS (long-term support) version from nodejs.org/en/download. On 2026-10-20 that was v24.22.0 (Node.js downloads).

git saves versions of your code. It's how you'll undo an agent's bad change in Module 04. Install it as the Git project describes. On macOS, Apple's Command Line Tools include git (or use Homebrew if you already have it). On Linux, use your package manager. On Windows, download "Git for Windows" or install it with winget. Open the list below and run the one command for your OS:

# macOS (pick one)
xcode-select --install        # Apple's Command Line Tools include git
brew install git              # if you already use Homebrew

# Debian / Ubuntu
sudo apt-get install git

# Fedora
sudo dnf install git

# Windows: download "Git for Windows" from git-scm.com/install/windows, or:
winget install --id Git.Git -e --source winget

Sources: macOS, Linux, Windows. On Windows, Git for Windows also matters for Claude Code. It provides Git Bash, which Claude Code's Bash tool uses. Without it, Claude Code falls back to PowerShell (Anthropic setup).

Close the terminal and open a new one so it picks up the new programs. Then check both:

node --version
git --version

Tell git who you are

Every commit records a name and email, and they're baked into the commit permanently (Pro Git §1.6). Set them once:

git config --global user.name "Your Name"
git config --global user.email "you@example.com"
git config --list    # check what git sees

Use the email on your GitHub account, so your commits link to your profile.

Laptop track, step 2: install and sign in to your agent

Install one agent, whichever you chose.

Setup A: install Claude Code

Anthropic's recommended native installer (Claude Code setup):

# macOS, Linux, WSL
curl -fsSL https://claude.ai/install.sh | bash

# Windows PowerShell (no Administrator needed)
irm https://claude.ai/install.ps1 | iex

The installer shows no progress while it downloads. Wait for it to finish, open a new terminal and run claude --version. A working install prints a version number (Claude Code setup). Claude Code supports macOS 13+, Windows 10 1809+, and Ubuntu 20.04+ / Debian 10+ (system requirements). On Windows it runs natively; WSL is optional (Claude Code setup).

Sign in: run claude. A browser window opens for you to log in with your Claude account. If it doesn't open, press c to copy the login URL and paste it into your browser. When the terminal shows Login successful, you're done (Authentication).

Setup B: install the Codex CLI

OpenAI's installers (Codex CLI, openai/codex README):

# macOS, Linux
curl -fsSL https://chatgpt.com/codex/install.sh | sh
# or, with Node already installed:
npm install -g @openai/codex

# Windows PowerShell
powershell -ExecutionPolicy ByPass -c "irm https://chatgpt.com/codex/install.ps1 | iex"

OpenAI documents native Windows support for the Codex CLI with a Windows sandbox, and recommends that sandbox by default over WSL (Codex on Windows).

Sign in: open a folder and run codex. The first time, choose Sign in with ChatGPT (Codex CLI). Signing in this way uses your plan. Signing in with an API key bills you at API rates instead (Codex auth), so pick ChatGPT sign-in for this course.

Grok Bot track: let a cloud agent install your coding CLI

In this setup, Grok Bot acts as the manager. Each Bot runs on a persistent cloud computer with a browser, filesystem and terminal, and it keeps working after you close your laptop (xAI overview). You ask it to install Claude Code or Codex in that terminal and drive it for you.

What you need before you start

• A Grok Bot–eligible plan. xAI's setup guide says you need a paid individual Cursor plan, the Cursor Teams plan, or an individual SuperGrok, SuperGrok Plus or SuperGrok Heavy subscription linked to your Cursor account (xAI get started). Cursor's plan page (checked 2026-10-20) lists Grok Bot on its Premium, Super and Ultra plans, but not on Free or Plus. It also says a SuperGrok or X Premium+ link grants usage only on a legacy Cursor account, and the link is permanent (Cursor plans). Plan names are in flux, so check that page before you buy.
• The Grok Bot desktop app, signed in with your Cursor account. Builds are available for macOS, Windows and Linux (xAI get started).
• A Claude plan (Pro or higher) or a ChatGPT plan for the coding CLI itself. Your Grok Bot plan doesn't sign you in to Claude or ChatGPT; you bring that login yourself.

Ask Grok Bot to install Claude Code or Codex

xAI's docs don't describe this workflow. They say the computer has a command line and a shared workspace at /workspace (xAI computer and apps). A third-party guide from MindStudio (October 2026) reports that the terminal behaves like a Linux machine, that you can tell a Bot to install Claude Code or Codex there, and that the install triggers the normal sign-in flow (MindStudio, third-party). This course's own runs match that: it's a Linux computer, and anything you install stays installed between sessions. Don't assume Node or git are already there; the prompt has the Bot check and install them. Create a Bot (one name, one job), then send it this:

Please set up a coding workspace in your terminal. Show me each command before you run it.
1. Tell me which Linux distribution this is (cat /etc/os-release).
   Then run `node --version` and `git --version`. If either is missing,
   install git with the system package manager and the current LTS
   version of Node.js the way nodejs.org/en/download describes for this
   distribution. Show me the commands first and wait for my OK.
2. Install Claude Code with Anthropic's official installer:
   curl -fsSL https://claude.ai/install.sh | bash
   (For Codex instead: curl -fsSL https://chatgpt.com/codex/install.sh | sh)
3. Create the folder /workspace/dayboard and start the CLI inside it.
4. When the CLI shows a sign-in link or asks for a code, send me only the link,
   then stop. I will take over the computer to finish signing in myself.
5. Never type passwords, tokens or codes into this chat.

Signing in safely. Claude Code may show a login code in your browser instead of redirecting back. That's common in containers, and you paste the code into the terminal (Anthropic authentication). For Codex, use device code login (codex login --device-auth), which you first enable in your ChatGPT security settings (Codex auth). Either way, use Grok Bot's takeover: open the computer, take control, complete only the blocked step, then tell the Bot to continue. xAI says to avoid pasting passwords or one-time codes into chat (xAI computer and apps).

Terms. Anthropic's Claude Code legal page says its rules don't prevent an end user from signing in to the unmodified Claude Code binary with their own Claude subscription, including where a platform hosts Claude Code (Anthropic legal and compliance). Anthropic's Consumer Terms also forbid sharing your login or making your account available to anyone else (Consumer Terms). OpenAI's Codex docs cover signing in with your ChatGPT account when "You’re running the CLI in a remote or headless environment", and recommend device code login for it (Codex auth). That's the situation on Grok Bot's computer. The same page warns: "Don't expose Codex execution in untrusted or public environments", which is one more reason to keep the Bot's computer to yourself.

Safety rules for a computer that isn't yours

xAI's FAQ is explicit: every Bot on your account uses one persistent cloud computer, with shared files, browser sessions, logins and command-line credentials, so separate Bots are not a security boundary (xAI FAQ, computer and apps). Deleting a Bot removes its profile, but files and logins on the computer may remain (xAI FAQ). xAI also says to treat manually installed packages as replaceable (xAI computer and apps). In this course's runs, installs survived between sessions, but if the CLI ever goes missing, ask the Bot to run the install prompt again.

Permission modes: who approves the agent's actions

A coding agent can edit files and run commands. Its permission mode decides which of those actions it pauses to ask you about. The defaults differ by tool, and they've changed recently:

Sources: Claude Code permission modes, Codex approvals and security, xAI security.

Anthropic warns that auto mode reduces prompts but doesn't guarantee safety (permission modes). For your first sessions, switch Claude Code to Manual (Shift+Tab once) so you see each request. That's the fastest way to learn what an agent actually does. xAI's advice holds for every tool: never approve an action whose target or effect you can't identify (xAI security). Lesson 01-02 explains the agent loop behind these prompts.

Hello, agent: verify your setup in an empty dayboard folder

Time to prove it works. Do it now.

1. Make the folder and start the agent there (laptop track):

mkdir dayboard
cd dayboard
claude          # or: codex

On the Grok Bot track, the Bot already made /workspace/dayboard and started the CLI there.

2. Set the permission mode. In Claude Code, press Shift+Tab until the footer shows Manual. In Codex, this folder isn't a git repo yet, so Codex recommends read-only, which can't write files (Codex approvals). Quit and restart it with the flags OpenAI documents as the Auto preset:

codex --sandbox workspace-write --ask-for-approval on-request

3. Paste this prompt. It works the same in every setup:

I'm starting a project called DayBoard: a personal morning dashboard that shows
my local weather next to my task list. Before we build anything, check the setup:
1. Run `node --version` and `git --version`, and report your own name and version
   (for example by running your CLI's --version command).
2. Create HELLO.md in this folder with those three versions, today's date, and
   the one-sentence description of DayBoard above.
3. Do not create or change any other file. Show me HELLO.md when you're done.

4. Watch and approve. Claude Code in Manual will ask before it runs commands or creates the file; read each request, then approve it. Codex in Auto may do both without asking, because they stay inside the folder, so read its transcript afterwards. On Grok Bot, ask it to paste the contents of HELLO.md into the chat.

What success looks like

# Hello, DayBoard
- Node.js: v24.22.0
- git: git version 2.56.0
- Agent: Claude Code 2.1.x   (or codex-cli x.y.z)
- Date: 2026-10-20

DayBoard is a personal morning dashboard that shows my local weather next to my task list.

Your numbers will differ. What matters is that three real version numbers appear and only one file exists in the folder. If the agent made extra files or invented a version without running the command, you've just met lesson 01-02's topic: agents can fail confidently. Tell it what it did wrong and ask it to fix the file.

What it costs to start (prices checked 2026-10-20)

The three platforms are free at DayBoard's scale. Your agent plan is the only real cost. No affiliate links. Every figure comes from the vendor's own pricing page.

So realistic starts are $20/month for setup A, $0–$20/month for setup B (Free and Go have tight limits), and about $50/month for setup C. Lesson 07-04 covers the full bill, including hosting once the app has real users.

Troubleshooting the common setup failures

Find your symptom in the right table. If nothing matches, paste the exact error text into the lesson chat. The exact wording is usually what points to the fix.

Install and PATH problems

Sign-in, permission and locked-down-laptop problems

Sources

• Anthropic — Claude Code docs: Advanced setup (2026)
• Anthropic — Claude Code docs: Terminal guide for new users (2026)
• Anthropic — Claude Code docs: Authentication (2026)
• Anthropic — Claude Code docs: Permission modes (2026)
• Anthropic — Claude Code docs: Legal and compliance (2026)
• Anthropic — Consumer Terms of Service (2026)
• Anthropic — Claude pricing (checked 2026-10-20)
• OpenAI — Codex CLI (2026)
• OpenAI — openai/codex README (2026)
• OpenAI — Codex authentication (2026)
• OpenAI — Codex agent approvals and security (2026)
• OpenAI — Codex on Windows: Windows sandbox (2026)
• OpenAI — ChatGPT Work and Codex pricing (checked 2026-10-20)
• xAI — Grok Bot overview (2026)
• xAI — Grok Bot: Get started (2026)
• xAI — Grok Bot: Computer and apps (2026)
• xAI — Grok Bot: Security (2026)
• xAI — Grok Bot: FAQ (2026)
• Cursor — Grok Bot plans and billing (checked 2026-10-20)
• Cursor — Pricing (checked 2026-10-20)
• MindStudio (third-party) — Run Claude Code and Codex inside Grok Bot (2026-10-09)
• GitHub Docs — Creating an account on GitHub (2026)
• GitHub Docs — Configuring two-factor authentication (2026)
• GitHub — Pricing (checked 2026-10-20)
• Vercel — Hobby plan docs (updated 2026-09-14)
• Vercel — Sign up (2026)
• Vercel — Getting started with Vercel (2026)
• Supabase — Pricing (checked 2026-10-20)
• Supabase — Dashboard sign-up (2026)
• OpenJS Foundation — Node.js downloads (checked 2026-10-20)
• Git project — Install Git (2026)
• Chacon & Straub — Pro Git, 2nd ed., §1.6 First-Time Git Setup (2014)

Practice

These steps start the DayBoard project you'll finish in the capstone.

1. Finish the checklist. Every row in the "Hello, agent" checklist should be ticked. If one isn't, use the troubleshooting table, or paste the exact error into the lesson chat.

2. Keep dayboard/HELLO.md. Don't delete it. You'll make it your first commit in lesson 04-01.

3. Start a notes file. Ask your agent to create dayboard/SETUP-NOTES.md that records:

   • which setup you chose (A, B or C) and why, in one sentence;
   • the exact Node.js, git and agent versions from HELLO.md;
   • the plan you're paying for and its monthly price today;
   • the permission mode you started in and how you switch modes in your tool.

   Read the file before you accept it. Does every version match what the commands actually printed?

4. Grok Bot track only: add a line to SETUP-NOTES.md listing every login now on the Bot's computer (for example "Claude account, via takeover"). You'll come back to this list when you add a GitHub token in lesson 04-02.

5. Stretch: start a fresh agent session in dayboard and ask: "Read HELLO.md and SETUP-NOTES.md and tell me if anything is inconsistent." That's your first taste of having the agent check work instead of only producing it.