Starter Guide

You don't need to code to run AI agents

From picking the right computer to setting up your terminal, this is the exact order we walk students through.

What's inside

01Free
Choosing a computer
Mac or Windows: set your starting criteria.
02Free
Understanding terminals
Why the terminal is your agent's office, with four tools introduced.
03Members only
Per-terminal setup
Setup recipes per tool and know-how for running agents in parallel.

PART 1

What makes a computer different in the agent era

An agent is like an employee who works alone for hours. It runs multiple sessions, spins up dev servers, and reads and writes files nonstop. Unlike chat-only AI use, memory capacity and OS fit strongly shape your throughput.

01

Memory equals parallel capacity

Running several agent sessions eats memory first. Apple Silicon Macs share one memory pool between CPU and GPU, which leaves more headroom for local AI work at the same capacity.

02

Sandboxed execution support

Claude Code's sandboxing works on macOS and Linux (WSL2). It's not yet supported on native Windows, so a Mac is easier if your workflow needs isolated runs.

03

Agent tools ship on Mac first

iTerm2 and cmux are Mac-only, and Ghostty supports macOS and Linux first. Most new agent-focused tools install with a single Homebrew command.

That said, Macs don't win everywhere. If you train models on NVIDIA GPUs, Linux is the better fit. This guide is written for agent operation and inference workloads.

Mac vs Windows at a glance

Item	Mac	Windows
Running Claude Code	Native support	Native support (PowerShell install)
Sandboxed execution	Supported	WSL2 only
Agent-focused terminals	iTerm2, cmux, Ghostty, and Orca all available	Mostly Orca and Windows Terminal
Tool installation	Unified via Homebrew	Install paths vary per tool

Good builds to buy now

A lineup picked for agent work. 16GB of memory is the starting line; go 24GB or more if you run many sessions in parallel.

MacBook Pro M5

Recommended

24GB+ memory recommended

Multiple agent sessions plus local servers: the safest choice as a main machine.

View on Coupang

MacBook Air M5

16GB+ memory

Plenty for getting started when you want to begin light.

View on Coupang

Mac mini M4

Desk setup

A desk build that buys you more memory for the same budget.

View on Coupang

Mac Studio in Silver with the M4 Max chip

Mac Studio M4 Max

36GB unified memory · 512GB SSD

Built for long parallel workloads and local models when you want one serious desk machine.

View on Coupang

Windows laptop

32GB+ RAM with WSL2

If you must stay on Windows, get generous memory and pair it with WSL2.

View on Coupang

This page contains Coupang Partners affiliate links; qualifying purchases may benefit QJC.

The Hermes agent that makes Claude Code 100x stronger — Real footage
Curious what an agent actually doing company work looks like? Start with this video.

PART 2

The terminal: your agent's office

What is a terminal?

A window where you talk to your computer in text. Instead of clicking icons, you type commands and the computer runs them as written. It looks foreign, but you only use a handful of rules daily.

Why does it matter now?

Because agents like Claude Code live and work inside the terminal. Once you're comfortable there, you gain a control room for running several agents at once.

iTerm2

macOS

The Mac terminal standard

A long-refined Mac-only terminal. Strong split panes and shortcuts make it great for keeping agent sessions side by side.

brew install --cask iterm2Official site

Ghostty

macOS · Linux

Fast, with minimal setup

Renders with the GPU, so it stays smooth even with heavy agent output. Sensible defaults let you use it right after install.

brew install --cask ghosttyOfficial site

cmux

macOS

Built for agent notifications

A terminal made for AI coding agents. When an agent waits for input, an indicator lights up, so you miss less while running many sessions.

brew tap manaflow-ai/cmux && brew install --cask cmuxOfficial site

Orca

macOS · Windows · Linux

Fleet control for agents

An orchestrator that runs multiple agents side by side, each isolated in its own working folder. A companion app lets you check progress from your phone.

brew install --cask stablyai/orca/orcaOfficial site

If you're on Windows

Claude Code installs natively on Windows too. Some features like sandboxing need WSL2, so check the setup criteria in Part 3.

Locked content preview

iTerm2
Ghostty
cmux
Orca
Shared rules for running agents in parallel

Part 3 · Students only

This part is open to our students

The small, annoying problems every terminal throws at you, we've already worked through them for you.

Setup recipes for all 4 terminals, click by click
Rules for running multiple agents in parallel
Content that keeps updating as the tools change

Explore Singularity Builders

Already a student? Log in to continue

You've made it this far. Next comes the real thing.

Singularity Builders is an AI agent bootcamp we run both in person and online. We'll go from setup to real work together.

Learn about Singularity Builders