Skip to content
Menu

Install your first AI agent

Pick the agent that fits the plan you already pay for, install it from the official instructions, and try it on something you can check.

Published October 25, 2026

The exact commands differ from one agent to the next, but the route is always the same. Each agent's page in the Library has the official commands, copied from the creator's own instructions; this guide covers the decisions around them.

  1. 1Pick the agentby the plan you have
  2. 2Install itofficial command
  3. 3Sign inor add a key
  4. 4Small first taskin a spare folder
  5. 5Check the result
The route from choosing to a first result

Start from what you already pay for

Every agent needs an AI model to think with, and the model is where the cost sits. So the easiest choice is usually the agent that works with an account you already have.

What do you already pay for?

  • A Claude plan (Pro, Max or a team plan)
    Claude logoClaude Code
  • A ChatGPT plan (Plus, Pro or a business plan)
    OpenAI logoCodex CLI
  • An API key, or you want to switch models
    OpenCode logogoose logoOpenCode or goose
  • Nothing yet, but you have a capable computer
    Ollama logoOpenCode or goose with Ollamaslower, and usually weaker answers
Which agent fits the plan you have

Running a model on your own computer through Ollama costs nothing, but you need a fairly powerful machine, and the answers are slower and usually not as good. It is a fine way to experiment; for real work most people end up on a plan or an API key.

Before you install, read "Before you start" on the agent's page. It lists the account it needs, the systems it runs on and anything else to set up first.

Install it

Most agents install with a single command in a terminal: Terminal on a Mac, PowerShell on Windows. On the agent's page, find the command for your system under "How to start", press "Copy install command", paste it into the terminal and press Enter. Some also offer a desktop app that installs like any other program.

If the command fails, resist trying fixes from random forum posts. When a setup fails walks through the usual causes.

Sign in

Start the agent by typing its name in the terminal (the page tells you the exact command). The first time, it asks you to sign in with your account or paste an API key. An API key works like a password, so paste it into the agent's own setup screen and nowhere else: not a chat, not a document.

Give it something small

Make a new folder, copy a few unimportant files into it, start the agent there, and ask for a job you can check yourself:

  • "List the files in this folder and tell me what each one is."
  • "Read notes.txt and write a five-line summary into summary.txt."
  • "Make a simple web page that shows today's date, and tell me how to open it."

Watch it work. It should say what it plans to do and ask before running commands or changing files. Answer each question until you know how it behaves. Then open the files it made. If something is off, say so in plain words; agents are good at fixing their own work once you point at the problem.

When that goes well, add a skill or a connector for a job you do often, and move on to real work one small task at a time.

Three things that catch people out

"Command not found" straight after installing usually means the terminal has not noticed the new program yet. Close it, open a new one and try again.

Some agents need a paid plan even for small tasks. The agent's page says so under "Before you start", so check there before you blame the install.

If it stops halfway through a long task, you may have hit your plan's usage limit for now. Wait a while, or break the job into smaller pieces.

Still stuck? Tell us where it stopped. We look at the problem first, then reply with a plan and a price.

Next: When a setup fails